Extract PG generation into pg_gen.js

.gitignore

@@ -3,16 +3,3 @@
 package-lock.json
 fio
 qemu
-osd
-stub_osd
-stub_uring_osd
-stub_bench
-osd_test
-osd_peering_pg_test
-dump_journal
-nbd_proxy
-rm_inode
-test_allocator
-test_blockstore
-test_shit
-osd_rmw_test

copy-fio-includes.sh

@@ -1,6 +1,6 @@
 #!/bin/bash
 
-gcc -I. -E -o fio_headers.i src/fio_headers.h
+gcc -I. -E -o fio_headers.i src/util/fio_headers.h
 
 rm -rf fio-copy
 for i in `grep -Po 'fio/[^"]+' fio_headers.i | sort | uniq`; do

copy-qemu-includes.sh

@@ -5,7 +5,7 @@
 #cd b/qemu; make qapi
 
 gcc -I qemu/b/qemu `pkg-config glib-2.0 --cflags` \
-    -I qemu/include -E -o qemu_driver.i src/qemu_driver.c
+    -I qemu/include -E -o qemu_driver.i src/client/qemu_driver.c
 
 rm -rf qemu-copy
 for i in `grep -Po 'qemu/[^"]+' qemu_driver.i | sort | uniq`; do

debian/patched-qemu.Dockerfile

@@ -27,7 +27,7 @@ RUN apt-get -y build-dep qemu
 RUN apt-get --download-only source qemu
 
 ADD patches /root/vitastor/patches
-ADD src/qemu_driver.c /root/vitastor/src/qemu_driver.c
+ADD src/client/qemu_driver.c /root/qemu_driver.c
 
 #RUN set -e; \
 #    apt-get install -y wget; \
@@ -52,7 +52,7 @@ RUN set -e; \
     cd /root/packages/qemu-$REL/qemu-*/; \
     quilt push -a; \
     quilt add block/vitastor.c; \
-    cp /root/vitastor/src/qemu_driver.c block/vitastor.c; \
+    cp /root/qemu_driver.c block/vitastor.c; \
     quilt refresh; \
     V=$(head -n1 debian/changelog | perl -pe 's/5\.2\+dfsg-9/5.2+dfsg-11/; s/^.*\((.*?)(~bpo[\d\+]*)?\).*$/$1/')+vitastor4; \
     if [ "$REL" = bullseye ]; then V=${V}bullseye; fi; \

debian/vitastor-mon.install

@@ -1,2 +1,3 @@
-mon usr/lib/vitastor
+mon usr/lib/vitastor/mon
-mon/vitastor-mon.service /lib/systemd/system
+mon/scripts/make-etcd usr/lib/vitastor/mon
+mon/scripts/vitastor-mon.service /lib/systemd/system

debian/vitastor-osd.install

@@ -1,6 +1,6 @@
 usr/bin/vitastor-osd
 usr/bin/vitastor-disk
 usr/bin/vitastor-dump-journal
-mon/vitastor-osd@.service /lib/systemd/system
+mon/scripts/vitastor-osd@.service /lib/systemd/system
-mon/vitastor.target /lib/systemd/system
+mon/scripts/vitastor.target /lib/systemd/system
-mon/90-vitastor.rules /lib/udev/rules.d
+mon/scripts/90-vitastor.rules /lib/udev/rules.d

docs/installation/source.en.md

@@ -41,7 +41,7 @@ It's recommended to build the QEMU driver (qemu_driver.c) in-tree, as a part of
 QEMU build process. To do that:
 - Install vitastor client library headers (from source or from vitastor-client-dev package)
 - Take a corresponding patch from `patches/qemu-*-vitastor.patch` and apply it to QEMU source
-- Copy `src/qemu_driver.c` to QEMU source directory as `block/vitastor.c`
+- Copy `src/client/qemu_driver.c` to QEMU source directory as `block/vitastor.c`
 - Build QEMU as usual
 
 But it is also possible to build it out-of-tree. To do that:

docs/installation/source.ru.md

@@ -41,7 +41,7 @@ cmake .. && make -j8 install
 Драйвер QEMU (qemu_driver.c) рекомендуется собирать вместе с самим QEMU. Для этого:
 - Установите заголовки клиентской библиотеки Vitastor (из исходников или из пакета vitastor-client-dev)
 - Возьмите соответствующий патч из `patches/qemu-*-vitastor.patch` и примените его к исходникам QEMU
-- Скопируйте [src/qemu_driver.c](../../src/qemu_driver.c) в директорию исходников QEMU как `block/vitastor.c`
+- Скопируйте [src/client/qemu_driver.c](../../src/client/qemu_driver.c) в директорию исходников QEMU как `block/vitastor.c`
 - Соберите QEMU как обычно
 
 Однако в целях отладки драйвер также можно собирать отдельно от QEMU. Для этого:

docs/intro/quickstart.en.md

@@ -22,7 +22,7 @@
   with lazy fsync, but prepare for inferior single-thread latency. Read more about capacitors
   [here](../config/layout-cluster.en.md#immediate_commit).
 - If you want to use HDDs, get modern HDDs with Media Cache or SSD Cache: HGST Ultrastar,
-  Toshiba MG08, Seagate EXOS or something similar. If your drives don't have such cache then
+  Toshiba MG, Seagate EXOS or something similar. If your drives don't have such cache then
   you also need small SSDs for journal and metadata (even 2 GB per 1 TB of HDD space is enough).
 - Get a fast network (at least 10 Gbit/s). Something like Mellanox ConnectX-4 with RoCEv2 is ideal.
 - Disable CPU powersaving: `cpupower idle-set -D 0 && cpupower frequency-set -g performance`.
@@ -33,7 +33,7 @@
 - SATA SSD: Micron 5100/5200/5300/5400, Samsung PM863/PM883/PM893, Intel D3-S4510/4520/4610/4620, Kingston DC500M
 - NVMe: Micron 9100/9200/9300/9400, Micron 7300/7450, Samsung PM983/PM9A3, Samsung PM1723/1735/1743,
   Intel DC-P3700/P4500/P4600, Intel D7-P5500/P5600, Intel Optane, Kingston DC1000B/DC1500M
-- HDD: HGST Ultrastar, Toshiba MG06/MG07/MG08, Seagate EXOS
+- HDD: HGST Ultrastar, Toshiba MG, Seagate EXOS
 
 ## Configure monitors
 

docs/intro/quickstart.ru.md

@@ -123,4 +123,4 @@ vitastor-cli create -s 10G testimg
 Если вы хотите использовать не только блочные образы виртуальных машин или контейнеров,
 а также кластерную файловую систему, то:
 
-- [Следуйте инструкциям](../usage/nfs.en.md#vitastorfs)
+- [Следуйте инструкциям](../usage/nfs.ru.md#vitastorfs)

mon/.eslintrc.js

@@ -11,6 +11,7 @@ module.exports = {
         "ecmaVersion": 2020
     },
     "plugins": [
+        "import"
     ],
     "rules": {
         "indent": [
@@ -44,6 +45,10 @@ module.exports = {
         ],
         "node/shebang": [
             "off"
+        ],
+        "import/no-unresolved": [
+            2,
+            { "commonjs": true }
         ]
     }
 };

mon/etcd_adapter.js

@@ -0,0 +1,320 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 (see README.md for details)
+
+const http = require('http');
+const WebSocket = require('ws');
+
+class EtcdAdapter
+{
+    constructor(mon)
+    {
+        this.mon = mon;
+        this.ws = null;
+        this.ws_alive = false;
+        this.ws_keepalive_timer = null;
+    }
+
+    parse_config(config)
+    {
+        this.parse_etcd_addresses(config.etcd_address||config.etcd_url);
+    }
+
+    parse_etcd_addresses(addrs)
+    {
+        const is_local_ip = this.mon.local_ips(true).reduce((a, c) => { a[c] = true; return a; }, {});
+        this.etcd_local = [];
+        this.etcd_urls = [];
+        this.selected_etcd_url = null;
+        this.etcd_urls_to_try = [];
+        if (!(addrs instanceof Array))
+            addrs = addrs ? (''+(addrs||'')).split(/,/) : [];
+        if (!addrs.length)
+        {
+            console.error('Vitastor etcd address(es) not specified. Please set on the command line or in the config file');
+            process.exit(1);
+        }
+        for (let url of addrs)
+        {
+            let scheme = 'http';
+            url = url.trim().replace(/^(https?):\/\//, (m, m1) => { scheme = m1; return ''; });
+            const slash = url.indexOf('/');
+            const colon = url.indexOf(':');
+            const is_local = is_local_ip[colon >= 0 ? url.substr(0, colon) : (slash >= 0 ? url.substr(0, slash) : url)];
+            url = scheme+'://'+(slash >= 0 ? url : url+'/v3');
+            if (is_local)
+                this.etcd_local.push(url);
+            else
+                this.etcd_urls.push(url);
+        }
+    }
+
+    pick_next_etcd()
+    {
+        if (this.selected_etcd_url)
+            return this.selected_etcd_url;
+        if (!this.etcd_urls_to_try || !this.etcd_urls_to_try.length)
+        {
+            this.etcd_urls_to_try = [ ...this.etcd_local ];
+            const others = [ ...this.etcd_urls ];
+            while (others.length)
+            {
+                const url = others.splice(0|(others.length*Math.random()), 1);
+                this.etcd_urls_to_try.push(url[0]);
+            }
+        }
+        this.selected_etcd_url = this.etcd_urls_to_try.shift();
+        return this.selected_etcd_url;
+    }
+
+    restart_watcher(cur_addr)
+    {
+        if (this.ws)
+        {
+            this.ws.close();
+            this.ws = null;
+        }
+        if (this.ws_keepalive_timer)
+        {
+            clearInterval(this.ws_keepalive_timer);
+            this.ws_keepalive_timer = null;
+        }
+        if (this.selected_etcd_url == cur_addr)
+        {
+            this.selected_etcd_url = null;
+        }
+        this.start_watcher(this.mon.config.etcd_mon_retries).catch(this.mon.die);
+    }
+
+    async start_watcher(retries)
+    {
+        let retry = 0;
+        if (!retries || retries < 1)
+        {
+            retries = 1;
+        }
+        const tried = {};
+        while (retries < 0 || retry < retries)
+        {
+            const cur_addr = this.pick_next_etcd();
+            const base = 'ws'+cur_addr.substr(4);
+            let now = Date.now();
+            if (tried[base] && now-tried[base] < this.mon.etcd_start_timeout)
+            {
+                await new Promise(ok => setTimeout(ok, this.mon.etcd_start_timeout-(now-tried[base])));
+                now = Date.now();
+            }
+            tried[base] = now;
+            const ok = await new Promise(ok =>
+            {
+                const timer_id = setTimeout(() =>
+                {
+                    this.ws.close();
+                    this.ws = null;
+                    ok(false);
+                }, this.mon.config.etcd_mon_timeout);
+                this.ws = new WebSocket(base+'/watch');
+                const fail = () =>
+                {
+                    ok(false);
+                };
+                this.ws.on('error', fail);
+                this.ws.on('open', () =>
+                {
+                    this.ws.removeListener('error', fail);
+                    if (timer_id)
+                        clearTimeout(timer_id);
+                    ok(true);
+                });
+            });
+            if (ok)
+                break;
+            if (this.selected_etcd_url == cur_addr)
+                this.selected_etcd_url = null;
+            this.ws = null;
+            retry++;
+        }
+        if (!this.ws)
+        {
+            this.mon.failconnect('Failed to open etcd watch websocket');
+        }
+        const cur_addr = this.selected_etcd_url;
+        this.ws_alive = true;
+        this.ws_keepalive_timer = setInterval(() =>
+        {
+            if (this.ws_alive)
+            {
+                this.ws_alive = false;
+                this.ws.send(JSON.stringify({ progress_request: {} }));
+            }
+            else
+            {
+                console.log('etcd websocket timed out, restarting it');
+                this.restart_watcher(cur_addr);
+            }
+        }, (Number(this.mon.config.etcd_ws_keepalive_interval) || 30)*1000);
+        this.ws.on('error', () => this.restart_watcher(cur_addr));
+        this.ws.send(JSON.stringify({
+            create_request: {
+                key: b64(this.mon.etcd_prefix+'/'),
+                range_end: b64(this.mon.etcd_prefix+'0'),
+                start_revision: ''+this.mon.etcd_watch_revision,
+                watch_id: 1,
+                progress_notify: true,
+            },
+        }));
+        this.ws.on('message', (msg) =>
+        {
+            this.ws_alive = true;
+            let data;
+            try
+            {
+                data = JSON.parse(msg);
+            }
+            catch (e)
+            {
+            }
+            if (!data || !data.result)
+            {
+                console.error('Unknown message received from watch websocket: '+msg);
+            }
+            else if (data.result.canceled)
+            {
+                // etcd watch canceled
+                if (data.result.compact_revision)
+                {
+                    // we may miss events if we proceed
+                    console.error('Revisions before '+data.result.compact_revision+' were compacted by etcd, exiting');
+                    this.mon.on_stop(1);
+                }
+                console.error('Watch canceled by etcd, reason: '+data.result.cancel_reason+', exiting');
+                this.mon.on_stop(1);
+            }
+            else if (data.result.created)
+            {
+                // etcd watch created
+            }
+            else
+            {
+                this.mon.on_message(data.result);
+            }
+        });
+    }
+
+    async become_master()
+    {
+        const state = { ...this.mon.get_mon_state(), id: ''+this.mon.etcd_lease_id };
+        // eslint-disable-next-line no-constant-condition
+        while (1)
+        {
+            const res = await this.etcd_call('/kv/txn', {
+                compare: [ { target: 'CREATE', create_revision: 0, key: b64(this.mon.etcd_prefix+'/mon/master') } ],
+                success: [ { requestPut: { key: b64(this.mon.etcd_prefix+'/mon/master'), value: b64(JSON.stringify(state)), lease: ''+this.mon.etcd_lease_id } } ],
+            }, this.mon.etcd_start_timeout, 0);
+            if (res.succeeded)
+            {
+                break;
+            }
+            console.log('Waiting to become master');
+            await new Promise(ok => setTimeout(ok, this.mon.etcd_start_timeout));
+        }
+        console.log('Became master');
+    }
+
+    async etcd_call(path, body, timeout, retries)
+    {
+        let retry = 0;
+        if (retries >= 0 && retries < 1)
+        {
+            retries = 1;
+        }
+        const tried = {};
+        while (retries < 0 || retry < retries)
+        {
+            retry++;
+            const base = this.pick_next_etcd();
+            let now = Date.now();
+            if (tried[base] && now-tried[base] < timeout)
+            {
+                await new Promise(ok => setTimeout(ok, timeout-(now-tried[base])));
+                now = Date.now();
+            }
+            tried[base] = now;
+            const res = await POST(base+path, body, timeout);
+            if (res.error)
+            {
+                if (this.selected_etcd_url == base)
+                    this.selected_etcd_url = null;
+                console.error('failed to query etcd: '+res.error);
+                continue;
+            }
+            if (res.json)
+            {
+                if (res.json.error)
+                {
+                    console.error('etcd returned error: '+res.json.error);
+                    break;
+                }
+                return res.json;
+            }
+        }
+        this.mon.failconnect();
+    }
+}
+
+function POST(url, body, timeout)
+{
+    return new Promise(ok =>
+    {
+        const body_text = Buffer.from(JSON.stringify(body));
+        let timer_id = timeout > 0 ? setTimeout(() =>
+        {
+            if (req)
+                req.abort();
+            req = null;
+            ok({ error: 'timeout' });
+        }, timeout) : null;
+        let req = http.request(url, { method: 'POST', headers: {
+            'Content-Type': 'application/json',
+            'Content-Length': body_text.length,
+        } }, (res) =>
+        {
+            if (!req)
+            {
+                return;
+            }
+            clearTimeout(timer_id);
+            let res_body = '';
+            res.setEncoding('utf8');
+            res.on('error', (error) => ok({ error }));
+            res.on('data', chunk => { res_body += chunk; });
+            res.on('end', () =>
+            {
+                if (res.statusCode != 200)
+                {
+                    ok({ error: res_body, code: res.statusCode });
+                    return;
+                }
+                try
+                {
+                    res_body = JSON.parse(res_body);
+                    ok({ response: res, json: res_body });
+                }
+                catch (e)
+                {
+                    ok({ error: e, response: res, body: res_body });
+                }
+            });
+        });
+        req.on('error', (error) => ok({ error }));
+        req.on('close', () => ok({ error: new Error('Connection closed prematurely') }));
+        req.write(body_text);
+        req.end();
+    });
+}
+
+function b64(str)
+{
+    return Buffer.from(str).toString('base64');
+}
+
+module.exports = EtcdAdapter;

mon/etcd_schema.js

@@ -0,0 +1,391 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 (see README.md for details)
+
+// FIXME document all etcd keys and config variables in the form of JSON schema or similar
+const etcd_nonempty_keys = {
+    'config/global': 1,
+    'config/node_placement': 1,
+    'config/pools': 1,
+    'config/pgs': 1,
+    'history/last_clean_pgs': 1,
+    'stats': 1,
+};
+const etcd_allow = new RegExp('^'+[
+    'config/global',
+    'config/node_placement',
+    'config/pools',
+    'config/osd/[1-9]\\d*',
+    'config/pgs',
+    'config/inode/[1-9]\\d*/[1-9]\\d*',
+    'osd/state/[1-9]\\d*',
+    'osd/stats/[1-9]\\d*',
+    'osd/inodestats/[1-9]\\d*',
+    'osd/space/[1-9]\\d*',
+    'mon/master',
+    'mon/member/[a-f0-9]+',
+    'pg/state/[1-9]\\d*/[1-9]\\d*',
+    'pg/stats/[1-9]\\d*/[1-9]\\d*',
+    'pg/history/[1-9]\\d*/[1-9]\\d*',
+    'history/last_clean_pgs',
+    'inode/stats/[1-9]\\d*/\\d+',
+    'pool/stats/[1-9]\\d*',
+    'stats',
+    'index/image/.*',
+    'index/maxid/[1-9]\\d*',
+].join('$|^')+'$');
+
+const etcd_tree = {
+    config: {
+        /* global: {
+            // WARNING: NOT ALL OF THESE ARE ACTUALLY CONFIGURABLE HERE
+            // THIS IS JUST A POOR MAN'S CONFIG DOCUMENTATION
+            // etcd connection
+            config_path: "/etc/vitastor/vitastor.conf",
+            etcd_prefix: "/vitastor",
+            // etcd connection - configurable online
+            etcd_address: "10.0.115.10:2379/v3",
+            // mon
+            etcd_mon_ttl: 5, // min: 1
+            etcd_mon_timeout: 1000, // ms. min: 0
+            etcd_mon_retries: 5, // min: 0
+            mon_change_timeout: 1000, // ms. min: 100
+            mon_retry_change_timeout: 50, // ms. min: 10
+            mon_stats_timeout: 1000, // ms. min: 100
+            osd_out_time: 600, // seconds. min: 0
+            placement_levels: { datacenter: 1, rack: 2, host: 3, osd: 4, ... },
+            use_old_pg_combinator: false,
+            // client and osd
+            tcp_header_buffer_size: 65536,
+            use_sync_send_recv: false,
+            use_rdma: true,
+            rdma_device: null, // for example, "rocep5s0f0"
+            rdma_port_num: 1,
+            rdma_gid_index: 0,
+            rdma_mtu: 4096,
+            rdma_max_sge: 128,
+            rdma_max_send: 8,
+            rdma_max_recv: 16,
+            rdma_max_msg: 132096,
+            block_size: 131072,
+            disk_alignment: 4096,
+            bitmap_granularity: 4096,
+            immediate_commit: false, // 'all' or 'small'
+            // client - configurable online
+            client_max_dirty_bytes: 33554432,
+            client_max_dirty_ops: 1024,
+            client_enable_writeback: false,
+            client_max_buffered_bytes: 33554432,
+            client_max_buffered_ops: 1024,
+            client_max_writeback_iodepth: 256,
+            client_retry_interval: 50, // ms. min: 10
+            client_eio_retry_interval: 1000, // ms
+            client_retry_enospc: true,
+            osd_nearfull_ratio: 0.95,
+            // client and osd - configurable online
+            log_level: 0,
+            peer_connect_interval: 5, // seconds. min: 1
+            peer_connect_timeout: 5, // seconds. min: 1
+            osd_idle_timeout: 5, // seconds. min: 1
+            osd_ping_timeout: 5, // seconds. min: 1
+            max_etcd_attempts: 5,
+            etcd_quick_timeout: 1000, // ms
+            etcd_slow_timeout: 5000, // ms
+            etcd_keepalive_timeout: 30, // seconds, default is max(30, etcd_report_interval*2)
+            etcd_ws_keepalive_interval: 30, // seconds
+            // osd
+            etcd_report_interval: 5, // seconds
+            etcd_stats_interval: 30, // seconds
+            run_primary: true,
+            osd_network: null, // "192.168.7.0/24" or an array of masks
+            bind_address: "0.0.0.0",
+            bind_port: 0,
+            readonly: false,
+            osd_memlock: false,
+            // osd - configurable online
+            autosync_interval: 5,
+            autosync_writes: 128,
+            client_queue_depth: 128, // unused
+            recovery_queue_depth: 1,
+            recovery_sleep_us: 0,
+            recovery_tune_util_low: 0.1,
+            recovery_tune_client_util_low: 0,
+            recovery_tune_util_high: 1.0,
+            recovery_tune_client_util_high: 0.5,
+            recovery_tune_interval: 1,
+            recovery_tune_agg_interval: 10, // 10 times recovery_tune_interval
+            recovery_tune_sleep_min_us: 10, // 10 microseconds
+            recovery_pg_switch: 128,
+            recovery_sync_batch: 16,
+            no_recovery: false,
+            no_rebalance: false,
+            print_stats_interval: 3,
+            slow_log_interval: 10,
+            inode_vanish_time: 60,
+            auto_scrub: false,
+            no_scrub: false,
+            scrub_interval: '30d', // 1s/1m/1h/1d
+            scrub_queue_depth: 1,
+            scrub_sleep: 0, // milliseconds
+            scrub_list_limit: 1000, // objects to list on one scrub iteration
+            scrub_find_best: true,
+            scrub_ec_max_bruteforce: 100, // maximum EC error locator brute-force iterators
+            // blockstore - fixed in superblock
+            block_size,
+            disk_alignment,
+            journal_block_size,
+            meta_block_size,
+            bitmap_granularity,
+            journal_device,
+            journal_offset,
+            journal_size,
+            disable_journal_fsync,
+            data_device,
+            data_offset,
+            data_size,
+            disable_data_fsync,
+            meta_device,
+            meta_offset,
+            disable_meta_fsync,
+            disable_device_lock,
+            // blockstore - configurable offline
+            inmemory_metadata,
+            inmemory_journal,
+            journal_sector_buffer_count,
+            journal_no_same_sector_overwrites,
+            // blockstore - configurable online
+            max_write_iodepth,
+            min_flusher_count: 1,
+            max_flusher_count: 256,
+            throttle_small_writes: false,
+            throttle_target_iops: 100,
+            throttle_target_mbs: 100,
+            throttle_target_parallelism: 1,
+            throttle_threshold_us: 50,
+        }, */
+        global: {},
+        /* node_placement: {
+            host1: { level: 'host', parent: 'rack1' },
+            ...
+        }, */
+        node_placement: {},
+        /* pools: {
+            <id>: {
+                name: 'testpool',
+                // 'ec' uses Reed-Solomon-Vandermonde codes, 'jerasure' is an alias for 'ec'
+                scheme: 'replicated' | 'xor' | 'ec' | 'jerasure',
+                pg_size: 3,
+                pg_minsize: 2,
+                // number of parity chunks, required for EC
+                parity_chunks?: 1,
+                pg_count: 100,
+                // default is failure_domain=host
+                failure_domain?: 'host',
+                // additional failure domain rules; failure_domain=x is equivalent to x=123..N
+                level_placement?: 'dc=112233 host=123456',
+                raw_placement?: 'any, dc=1 host!=1, dc=1 host!=(1,2)',
+                old_combinator: false,
+                max_osd_combinations: 10000,
+                // block_size, bitmap_granularity, immediate_commit must match all OSDs used in that pool
+                block_size: 131072,
+                bitmap_granularity: 4096,
+                // 'all'/'small'/'none', same as in OSD options
+                immediate_commit: 'none',
+                pg_stripe_size: 0,
+                root_node?: 'rack1',
+                // restrict pool to OSDs having all of these tags
+                osd_tags?: 'nvme' | [ 'nvme', ... ],
+                // prefer to put primary on OSD with these tags
+                primary_affinity_tags?: 'nvme' | [ 'nvme', ... ],
+                // scrub interval
+                scrub_interval?: '30d',
+            },
+            ...
+        }, */
+        pools: {},
+        osd: {
+            /* <id>: { reweight?: 1, tags?: [ 'nvme', ... ], noout?: true }, ... */
+        },
+        /* pgs: {
+            hash: string,
+            items: {
+                <pool_id>: {
+                    <pg_id>: {
+                        osd_set: [ 1, 2, 3 ],
+                        primary: 1,
+                        pause: false,
+                    }
+                }
+            }
+        }, */
+        pgs: {},
+        /* inode: {
+            <pool_id>: {
+                <inode_t>: {
+                    name: string,
+                    size?: uint64_t, // bytes
+                    parent_pool?: <pool_id>,
+                    parent_id?: <inode_t>,
+                    readonly?: boolean,
+                }
+            }
+        }, */
+        inode: {},
+    },
+    osd: {
+        state: {
+            /* <osd_num_t>: {
+                state: "up",
+                addresses: string[],
+                host: string,
+                port: uint16_t,
+                primary_enabled: boolean,
+                blockstore_enabled: boolean,
+            }, */
+        },
+        stats: {
+            /* <osd_num_t>: {
+                time: number, // unix time
+                blockstore_ready: boolean,
+                size: uint64_t, // bytes
+                free: uint64_t, // bytes
+                host: string,
+                op_stats: {
+                    <string>: { count: uint64_t, usec: uint64_t, bytes: uint64_t },
+                },
+                subop_stats: {
+                    <string>: { count: uint64_t, usec: uint64_t },
+                },
+                recovery_stats: {
+                    degraded: { count: uint64_t, bytes: uint64_t },
+                    misplaced: { count: uint64_t, bytes: uint64_t },
+                },
+            }, */
+        },
+        inodestats: {
+            /* <pool_id>: {
+                <inode_t>: {
+                    read: { count: uint64_t, usec: uint64_t, bytes: uint64_t },
+                    write: { count: uint64_t, usec: uint64_t, bytes: uint64_t },
+                    delete: { count: uint64_t, usec: uint64_t, bytes: uint64_t },
+                },
+            }, */
+        },
+        space: {
+            /* <osd_num_t>: {
+                <pool_id>: {
+                    <inode_t>: uint64_t, // bytes
+                },
+            }, */
+        },
+    },
+    mon: {
+        master: {
+            /* ip: [ string ], id: uint64_t */
+        },
+        standby: {
+            /* <uint64_t>: { ip: [ string ] }, */
+        },
+    },
+    pg: {
+        state: {
+            /* <pool_id>: {
+                <pg_id>: {
+                    primary: osd_num_t,
+                    state: ("starting"|"peering"|"incomplete"|"active"|"repeering"|"stopping"|"offline"|
+                        "degraded"|"has_incomplete"|"has_degraded"|"has_misplaced"|"has_unclean"|
+                        "has_invalid"|"has_inconsistent"|"has_corrupted"|"left_on_dead"|"scrubbing")[],
+                }
+            }, */
+        },
+        stats: {
+            /* <pool_id>: {
+                <pg_id>: {
+                    object_count: uint64_t,
+                    clean_count: uint64_t,
+                    misplaced_count: uint64_t,
+                    degraded_count: uint64_t,
+                    incomplete_count: uint64_t,
+                    write_osd_set: osd_num_t[],
+                },
+            }, */
+        },
+        history: {
+            /* <pool_id>: {
+                <pg_id>: {
+                    osd_sets: osd_num_t[][],
+                    all_peers: osd_num_t[],
+                    epoch: uint64_t,
+                    next_scrub: uint64_t,
+                },
+            }, */
+        },
+    },
+    inode: {
+        stats: {
+            /* <pool_id>: {
+                <inode_t>: {
+                    raw_used: uint64_t, // raw used bytes on OSDs
+                    read: { count: uint64_t, usec: uint64_t, bytes: uint64_t, bps: uint64_t, iops: uint64_t, lat: uint64_t },
+                    write: { count: uint64_t, usec: uint64_t, bytes: uint64_t, bps: uint64_t, iops: uint64_t, lat: uint64_t },
+                    delete: { count: uint64_t, usec: uint64_t, bytes: uint64_t, bps: uint64_t, iops: uint64_t, lat: uint64_t },
+                },
+            }, */
+        },
+    },
+    pool: {
+        stats: {
+            /* <pool_id>: {
+                used_raw_tb: float, // used raw space in the pool
+                total_raw_tb: float, // maximum amount of space in the pool
+                raw_to_usable: float, // raw to usable ratio
+                space_efficiency: float, // 0..1
+            } */
+        },
+    },
+    stats: {
+        /* op_stats: {
+            <string>: { count: uint64_t, usec: uint64_t, bytes: uint64_t, bps: uint64_t, iops: uint64_t, lat: uint64_t },
+        },
+        subop_stats: {
+            <string>: { count: uint64_t, usec: uint64_t, iops: uint64_t, lat: uint64_t },
+        },
+        recovery_stats: {
+            degraded: { count: uint64_t, bytes: uint64_t, bps: uint64_t, iops: uint64_t },
+            misplaced: { count: uint64_t, bytes: uint64_t, bps: uint64_t, iops: uint64_t },
+        },
+        object_counts: {
+            object: uint64_t,
+            clean: uint64_t,
+            misplaced: uint64_t,
+            degraded: uint64_t,
+            incomplete: uint64_t,
+        },
+        object_bytes: {
+            total: uint64_t,
+            clean: uint64_t,
+            misplaced: uint64_t,
+            degraded: uint64_t,
+            incomplete: uint64_t,
+        }, */
+    },
+    history: {
+        last_clean_pgs: {},
+    },
+    index: {
+        image: {
+            /* <name>: {
+                id: uint64_t,
+                pool_id: uint64_t,
+            }, */
+        },
+        maxid: {
+            /* <pool_id>: uint64_t, */
+        },
+    },
+};
+
+module.exports = {
+    etcd_nonempty_keys,
+    etcd_allow,
+    etcd_tree,
+};

mon/lp_optimizer/test-nonuniform.js

@@ -8,7 +8,7 @@
 // But we support this case with the "parity_space" parameter in optimize_initial()/optimize_change().
 
 const { SimpleCombinator } = require('./simple_pgs.js');
-const LPOptimizer = require('./lp-optimizer.js');
+const LPOptimizer = require('./lp_optimizer.js');
 
 const osd_tree = {
     ripper5: {

mon/lp_optimizer/test-optimize-simple.js

@@ -2,7 +2,7 @@
 // License: VNPL-1.1 (see README.md for details)
 
 const { compat } = require('./simple_pgs.js');
-const LPOptimizer = require('./lp-optimizer.js');
+const LPOptimizer = require('./lp_optimizer.js');
 
 async function run()
 {

mon/lp_optimizer/test-optimize-undersized.js

@@ -2,7 +2,7 @@
 // License: VNPL-1.1 (see README.md for details)
 
 const { compat, flatten_tree } = require('./simple_pgs.js');
-const LPOptimizer = require('./lp-optimizer.js');
+const LPOptimizer = require('./lp_optimizer.js');
 
 const crush_tree = [
     { level: 1, children: [

mon/lp_optimizer/test-optimize-unfeasible.js

@@ -2,7 +2,7 @@
 // License: VNPL-1.1 (see README.md for details)
 
 const { compat } = require('./simple_pgs.js');
-const LPOptimizer = require('./lp-optimizer.js');
+const LPOptimizer = require('./lp_optimizer.js');
 
 const osd_tree = {
     100: {

mon/lp_optimizer/test-optimize.js

@@ -2,7 +2,7 @@
 // License: VNPL-1.1 (see README.md for details)
 
 const { compat, flatten_tree } = require('./simple_pgs.js');
-const LPOptimizer = require('./lp-optimizer.js');
+const LPOptimizer = require('./lp_optimizer.js');
 
 const osd_tree = {
     100: {

mon/osd_tree.js

@@ -0,0 +1,215 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 (see README.md for details)
+
+function get_osd_tree(global_config, state)
+{
+    const levels = global_config.placement_levels||{};
+    levels.host = levels.host || 100;
+    levels.osd = levels.osd || 101;
+    const tree = {};
+    let up_osds = {};
+    // This requires monitor system time to be in sync with OSD system times (at least to some extent)
+    const down_time = Date.now()/1000 - global_config.osd_out_time;
+    for (const osd_num of Object.keys(state.osd.stats).sort((a, b) => a - b))
+    {
+        const stat = state.osd.stats[osd_num];
+        const osd_cfg = state.config.osd[osd_num];
+        let reweight = osd_cfg == null ? 1 : Number(osd_cfg.reweight);
+        if (reweight < 0 || isNaN(reweight))
+            reweight = 1;
+        if (stat && stat.size && reweight && (state.osd.state[osd_num] || Number(stat.time) >= down_time ||
+            osd_cfg && osd_cfg.noout))
+        {
+            // Numeric IDs are reserved for OSDs
+            if (state.osd.state[osd_num] && reweight > 0)
+            {
+                // React to down OSDs immediately
+                up_osds[osd_num] = true;
+            }
+            tree[osd_num] = tree[osd_num] || {};
+            tree[osd_num].id = osd_num;
+            tree[osd_num].parent = tree[osd_num].parent || stat.host;
+            tree[osd_num].level = 'osd';
+            tree[osd_num].size = reweight * stat.size / 1024 / 1024 / 1024 / 1024; // terabytes
+            if (osd_cfg && osd_cfg.tags)
+            {
+                tree[osd_num].tags = (osd_cfg.tags instanceof Array ? [ ...osd_cfg.tags ] : [ osd_cfg.tags ])
+                    .reduce((a, c) => { a[c] = true; return a; }, {});
+            }
+            delete tree[osd_num].children;
+            if (!tree[stat.host])
+            {
+                tree[stat.host] = {
+                    id: stat.host,
+                    level: 'host',
+                    parent: null,
+                    children: [],
+                };
+            }
+        }
+    }
+    for (const node_id in state.config.node_placement||{})
+    {
+        const node_cfg = state.config.node_placement[node_id];
+        if (/^\d+$/.exec(node_id))
+        {
+            node_cfg.level = 'osd';
+        }
+        if (!node_id || !node_cfg.level || !levels[node_cfg.level] ||
+            node_cfg.level === 'osd' && !tree[node_id])
+        {
+            // All nodes must have non-empty IDs and valid levels
+            // OSDs have to actually exist
+            continue;
+        }
+        tree[node_id] = tree[node_id] || {};
+        tree[node_id].id = node_id;
+        tree[node_id].level = node_cfg.level;
+        tree[node_id].parent = node_cfg.parent;
+        if (node_cfg.level !== 'osd')
+        {
+            tree[node_id].children = [];
+        }
+    }
+    return { up_osds, levels, osd_tree: tree };
+}
+
+function make_hier_tree(global_config, tree)
+{
+    const levels = global_config.placement_levels||{};
+    levels.host = levels.host || 100;
+    levels.osd = levels.osd || 101;
+    tree = { ...tree };
+    for (const node_id in tree)
+    {
+        tree[node_id] = { ...tree[node_id], children: [] };
+    }
+    tree[''] = { children: [] };
+    for (const node_id in tree)
+    {
+        if (node_id === '' || tree[node_id].level === 'osd' && (!tree[node_id].size || tree[node_id].size <= 0))
+        {
+            continue;
+        }
+        const node_cfg = tree[node_id];
+        const node_level = levels[node_cfg.level] || node_cfg.level;
+        let parent_level = node_cfg.parent && tree[node_cfg.parent] && tree[node_cfg.parent].children
+            && tree[node_cfg.parent].level;
+        parent_level = parent_level ? (levels[parent_level] || parent_level) : null;
+        // Parent's level must be less than child's; OSDs must be leaves
+        const parent = parent_level && parent_level < node_level ? node_cfg.parent : '';
+        tree[parent].children.push(tree[node_id]);
+    }
+    // Delete empty nodes
+    let deleted = 0;
+    do
+    {
+        deleted = 0;
+        for (const node_id in tree)
+        {
+            if (tree[node_id].level !== 'osd' && (!tree[node_id].children || !tree[node_id].children.length))
+            {
+                const parent = tree[node_id].parent;
+                if (parent)
+                {
+                    tree[parent].children = tree[parent].children.filter(c => c != tree[node_id]);
+                }
+                deleted++;
+                delete tree[node_id];
+            }
+        }
+    } while (deleted > 0);
+    return tree;
+}
+
+function filter_osds_by_root_node(global_config, pool_tree, root_node)
+{
+    if (!root_node)
+    {
+        return;
+    }
+    let hier_tree = make_hier_tree(global_config, pool_tree);
+    let included = [ ...(hier_tree[root_node] || {}).children||[] ];
+    for (let i = 0; i < included.length; i++)
+    {
+        if (included[i].children)
+        {
+            included.splice(i+1, 0, ...included[i].children);
+        }
+    }
+    let cur = pool_tree[root_node] || {};
+    while (cur && cur.id)
+    {
+        included.unshift(cur);
+        cur = pool_tree[cur.parent||''];
+    }
+    included = included.reduce((a, c) => { a[c.id||''] = true; return a; }, {});
+    for (const item in pool_tree)
+    {
+        if (!included[item])
+        {
+            delete pool_tree[item];
+        }
+    }
+}
+
+function filter_osds_by_tags(orig_tree, tags)
+{
+    if (!tags)
+    {
+        return;
+    }
+    for (const tag of (tags instanceof Array ? tags : [ tags ]))
+    {
+        for (const osd in orig_tree)
+        {
+            if (orig_tree[osd].level === 'osd' &&
+                (!orig_tree[osd].tags || !orig_tree[osd].tags[tag]))
+            {
+                delete orig_tree[osd];
+            }
+        }
+    }
+}
+
+function filter_osds_by_block_layout(orig_tree, osd_stats, block_size, bitmap_granularity, immediate_commit)
+{
+    for (const osd in orig_tree)
+    {
+        if (orig_tree[osd].level === 'osd')
+        {
+            const osd_stat = osd_stats[osd];
+            if (osd_stat && (osd_stat.bs_block_size && osd_stat.bs_block_size != block_size ||
+                osd_stat.bitmap_granularity && osd_stat.bitmap_granularity != bitmap_granularity ||
+                osd_stat.immediate_commit == 'small' && immediate_commit == 'all' ||
+                osd_stat.immediate_commit == 'none' && immediate_commit != 'none'))
+            {
+                delete orig_tree[osd];
+            }
+        }
+    }
+}
+
+function get_affinity_osds(pool_cfg, up_osds, osd_tree)
+{
+    let aff_osds = up_osds;
+    if (pool_cfg.primary_affinity_tags)
+    {
+        aff_osds = Object.keys(up_osds).reduce((a, c) => { a[c] = osd_tree[c]; return a; }, {});
+        filter_osds_by_tags(aff_osds, pool_cfg.primary_affinity_tags);
+        for (const osd in aff_osds)
+        {
+            aff_osds[osd] = true;
+        }
+    }
+    return aff_osds;
+}
+
+module.exports = {
+    get_osd_tree,
+    make_hier_tree,
+    filter_osds_by_root_node,
+    filter_osds_by_tags,
+    filter_osds_by_block_layout,
+    get_affinity_osds,
+};

mon/package.json

@@ -4,7 +4,7 @@
   "description": "Vitastor SDS monitor service",
   "main": "mon-main.js",
   "scripts": {
-    "test": "echo \"Error: no test specified\" && exit 1"
+    "lint": "eslint *.js lp_optimizer/*.js scripts/*.js"
   },
   "author": "Vitaliy Filippov",
   "license": "UNLICENSED",
@@ -14,12 +14,10 @@
   },
   "devDependencies": {
     "eslint": "^8.0.0",
+    "eslint-plugin-import": "^2.29.1",
     "eslint-plugin-node": "^11.1.0"
   },
   "engines": {
     "node": ">=12.0.0"
-  },
-  "scripts": {
-    "lint": "eslint *.js"
   }
 }

mon/pg_gen.js

@@ -0,0 +1,267 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 (see README.md for details)
+
+const { RuleCombinator } = require('./lp_optimizer/dsl_pgs.js');
+const { SimpleCombinator, flatten_tree } = require('./lp_optimizer/simple_pgs.js');
+const { validate_pool_cfg, get_pg_rules } = require('./pool_config.js');
+const LPOptimizer = require('./lp_optimizer/lp_optimizer.js');
+const { scale_pg_count } = require('./pg_utils.js');
+const { make_hier_tree, filter_osds_by_root_node,
+    filter_osds_by_tags, filter_osds_by_block_layout, get_affinity_osds } = require('./osd_tree.js');
+
+let seed;
+
+function reset_rng()
+{
+    seed = 0x5f020e43;
+}
+
+function rng()
+{
+    seed ^= seed << 13;
+    seed ^= seed >> 17;
+    seed ^= seed << 5;
+    return seed + 2147483648;
+}
+
+function pick_primary(pool_config, osd_set, up_osds, aff_osds)
+{
+    let alive_set;
+    if (pool_config.scheme === 'replicated')
+    {
+        // Prefer "affinity" OSDs
+        alive_set = osd_set.filter(osd_num => osd_num && aff_osds[osd_num]);
+        if (!alive_set.length)
+            alive_set = osd_set.filter(osd_num => osd_num && up_osds[osd_num]);
+    }
+    else
+    {
+        // Prefer data OSDs for EC because they can actually read something without an additional network hop
+        const pg_data_size = (pool_config.pg_size||0) - (pool_config.parity_chunks||0);
+        alive_set = osd_set.slice(0, pg_data_size).filter(osd_num => osd_num && aff_osds[osd_num]);
+        if (!alive_set.length)
+            alive_set = osd_set.filter(osd_num => osd_num && aff_osds[osd_num]);
+        if (!alive_set.length)
+        {
+            alive_set = osd_set.slice(0, pg_data_size).filter(osd_num => osd_num && up_osds[osd_num]);
+            if (!alive_set.length)
+                alive_set = osd_set.filter(osd_num => osd_num && up_osds[osd_num]);
+        }
+    }
+    if (!alive_set.length)
+    {
+        return 0;
+    }
+    return alive_set[rng() % alive_set.length];
+}
+
+function recheck_primary(state, global_config, up_osds, osd_tree)
+{
+    let new_config_pgs;
+    for (const pool_id in state.config.pools)
+    {
+        const pool_cfg = state.config.pools[pool_id];
+        if (!validate_pool_cfg(pool_id, pool_cfg, global_config.placement_levels, false))
+        {
+            continue;
+        }
+        const aff_osds = get_affinity_osds(pool_cfg, up_osds, osd_tree);
+        reset_rng();
+        for (let pg_num = 1; pg_num <= pool_cfg.pg_count; pg_num++)
+        {
+            if (!state.config.pgs.items[pool_id])
+            {
+                continue;
+            }
+            const pg_cfg = state.config.pgs.items[pool_id][pg_num];
+            if (pg_cfg)
+            {
+                const new_primary = pick_primary(state.config.pools[pool_id], pg_cfg.osd_set, up_osds, aff_osds);
+                if (pg_cfg.primary != new_primary)
+                {
+                    if (!new_config_pgs)
+                    {
+                        new_config_pgs = JSON.parse(JSON.stringify(state.config.pgs));
+                    }
+                    console.log(
+                        `Moving pool ${pool_id} (${pool_cfg.name || 'unnamed'}) PG ${pg_num}`+
+                        ` primary OSD from ${pg_cfg.primary} to ${new_primary}`
+                    );
+                    new_config_pgs.items[pool_id][pg_num].primary = new_primary;
+                }
+            }
+        }
+    }
+    return new_config_pgs;
+}
+
+function save_new_pgs_txn(save_to, request, state, etcd_prefix, etcd_watch_revision, pool_id, up_osds, osd_tree, prev_pgs, new_pgs, pg_history)
+{
+    const aff_osds = get_affinity_osds(state.config.pools[pool_id] || {}, up_osds, osd_tree);
+    const pg_items = {};
+    reset_rng();
+    new_pgs.map((osd_set, i) =>
+    {
+        osd_set = osd_set.map(osd_num => osd_num === LPOptimizer.NO_OSD ? 0 : osd_num);
+        pg_items[i+1] = {
+            osd_set,
+            primary: pick_primary(state.config.pools[pool_id], osd_set, up_osds, aff_osds),
+        };
+        if (prev_pgs[i] && prev_pgs[i].join(' ') != osd_set.join(' ') &&
+            prev_pgs[i].filter(osd_num => osd_num).length > 0)
+        {
+            pg_history[i] = pg_history[i] || {};
+            pg_history[i].osd_sets = pg_history[i].osd_sets || [];
+            pg_history[i].osd_sets.push(prev_pgs[i]);
+        }
+        if (pg_history[i] && pg_history[i].osd_sets)
+        {
+            pg_history[i].osd_sets = Object.values(pg_history[i].osd_sets
+                .reduce((a, c) => { a[c.join(' ')] = c; return a; }, {}));
+        }
+    });
+    for (let i = 0; i < new_pgs.length || i < prev_pgs.length; i++)
+    {
+        // FIXME: etcd has max_txn_ops limit, and it's 128 by default
+        // Sooo we probably want to change our storage scheme for PG histories...
+        request.compare.push({
+            key: b64(etcd_prefix+'/pg/history/'+pool_id+'/'+(i+1)),
+            target: 'MOD',
+            mod_revision: ''+etcd_watch_revision,
+            result: 'LESS',
+        });
+        if (pg_history[i])
+        {
+            request.success.push({
+                requestPut: {
+                    key: b64(etcd_prefix+'/pg/history/'+pool_id+'/'+(i+1)),
+                    value: b64(JSON.stringify(pg_history[i])),
+                },
+            });
+        }
+        else
+        {
+            request.success.push({
+                requestDeleteRange: {
+                    key: b64(etcd_prefix+'/pg/history/'+pool_id+'/'+(i+1)),
+                },
+            });
+        }
+    }
+    save_to.items = save_to.items || {};
+    if (!new_pgs.length)
+    {
+        delete save_to.items[pool_id];
+    }
+    else
+    {
+        save_to.items[pool_id] = pg_items;
+    }
+}
+
+async function generate_pool_pgs(state, global_config, pool_id, osd_tree, levels)
+{
+    const pool_cfg = state.config.pools[pool_id];
+    if (!validate_pool_cfg(pool_id, pool_cfg, global_config.placement_levels, false))
+    {
+        return null;
+    }
+    let pool_tree = { ...osd_tree };
+    filter_osds_by_root_node(global_config, pool_tree, pool_cfg.root_node);
+    filter_osds_by_tags(pool_tree, pool_cfg.osd_tags);
+    filter_osds_by_block_layout(
+        pool_tree,
+        state.osd.stats,
+        pool_cfg.block_size || global_config.block_size || 131072,
+        pool_cfg.bitmap_granularity || global_config.bitmap_granularity || 4096,
+        pool_cfg.immediate_commit || global_config.immediate_commit || 'none'
+    );
+    pool_tree = make_hier_tree(global_config, pool_tree);
+    // First try last_clean_pgs to minimize data movement
+    let prev_pgs = [];
+    for (const pg in ((state.history.last_clean_pgs.items||{})[pool_id]||{}))
+    {
+        prev_pgs[pg-1] = [ ...state.history.last_clean_pgs.items[pool_id][pg].osd_set ];
+    }
+    if (!prev_pgs.length)
+    {
+        // Fall back to config/pgs if it's empty
+        for (const pg in ((state.config.pgs.items||{})[pool_id]||{}))
+        {
+            prev_pgs[pg-1] = [ ...state.config.pgs.items[pool_id][pg].osd_set ];
+        }
+    }
+    const old_pg_count = prev_pgs.length;
+    const optimize_cfg = {
+        osd_weights: Object.values(pool_tree).filter(item => item.level === 'osd').reduce((a, c) => { a[c.id] = c.size; return a; }, {}),
+        combinator: !global_config.use_old_pg_combinator || pool_cfg.level_placement || pool_cfg.raw_placement
+            // new algorithm:
+            ? new RuleCombinator(pool_tree, get_pg_rules(pool_id, pool_cfg, global_config.placement_levels), pool_cfg.max_osd_combinations)
+            // old algorithm:
+            : new SimpleCombinator(flatten_tree(pool_tree[''].children, levels, pool_cfg.failure_domain, 'osd'), pool_cfg.pg_size, pool_cfg.max_osd_combinations),
+        pg_count: pool_cfg.pg_count,
+        pg_size: pool_cfg.pg_size,
+        pg_minsize: pool_cfg.pg_minsize,
+        ordered: pool_cfg.scheme != 'replicated',
+    };
+    let optimize_result;
+    // Re-shuffle PGs if config/pgs.hash is empty
+    if (old_pg_count > 0 && state.config.pgs.hash)
+    {
+        if (prev_pgs.length != pool_cfg.pg_count)
+        {
+            // Scale PG count
+            // Do it even if old_pg_count is already equal to pool_cfg.pg_count,
+            // because last_clean_pgs may still contain the old number of PGs
+            scale_pg_count(prev_pgs, pool_cfg.pg_count);
+        }
+        for (const pg of prev_pgs)
+        {
+            while (pg.length < pool_cfg.pg_size)
+            {
+                pg.push(0);
+            }
+        }
+        optimize_result = await LPOptimizer.optimize_change({
+            prev_pgs,
+            ...optimize_cfg,
+        });
+    }
+    else
+    {
+        optimize_result = await LPOptimizer.optimize_initial(optimize_cfg);
+    }
+    console.log(`Pool ${pool_id} (${pool_cfg.name || 'unnamed'}):`);
+    LPOptimizer.print_change_stats(optimize_result);
+    let pg_effsize = pool_cfg.pg_size;
+    for (const pg of optimize_result.int_pgs)
+    {
+        const this_pg_size = pg.filter(osd => osd != LPOptimizer.NO_OSD).length;
+        if (this_pg_size && this_pg_size < pg_effsize)
+        {
+            pg_effsize = this_pg_size;
+        }
+    }
+    return {
+        pool_id,
+        pgs: optimize_result.int_pgs,
+        stats: {
+            total_raw_tb: optimize_result.space,
+            pg_real_size: pg_effsize || pool_cfg.pg_size,
+            raw_to_usable: (pg_effsize || pool_cfg.pg_size) / (pool_cfg.scheme === 'replicated'
+                ? 1 : (pool_cfg.pg_size - (pool_cfg.parity_chunks||0))),
+            space_efficiency: optimize_result.space/(optimize_result.total_space||1),
+        },
+    };
+}
+
+function b64(str)
+{
+    return Buffer.from(str).toString('base64');
+}
+
+module.exports = {
+    recheck_primary,
+    save_new_pgs_txn,
+    generate_pool_pgs,
+};

mon/pool_config.js

@@ -0,0 +1,169 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 (see README.md for details)
+
+const { parse_level_indexes, parse_pg_dsl } = require('./lp_optimizer/dsl_pgs.js');
+
+function validate_pool_cfg(pool_id, pool_cfg, placement_levels, warn)
+{
+    pool_cfg.pg_size = Math.floor(pool_cfg.pg_size);
+    pool_cfg.pg_minsize = Math.floor(pool_cfg.pg_minsize);
+    pool_cfg.parity_chunks = Math.floor(pool_cfg.parity_chunks) || undefined;
+    pool_cfg.pg_count = Math.floor(pool_cfg.pg_count);
+    pool_cfg.max_osd_combinations = Math.floor(pool_cfg.max_osd_combinations) || 10000;
+    if (!/^[1-9]\d*$/.exec(''+pool_id))
+    {
+        if (warn)
+            console.log('Pool ID '+pool_id+' is invalid');
+        return false;
+    }
+    if (pool_cfg.scheme !== 'xor' && pool_cfg.scheme !== 'replicated' &&
+        pool_cfg.scheme !== 'ec' && pool_cfg.scheme !== 'jerasure')
+    {
+        if (warn)
+            console.log('Pool '+pool_id+' has invalid coding scheme (one of "xor", "replicated", "ec" and "jerasure" required)');
+        return false;
+    }
+    if (!pool_cfg.pg_size || pool_cfg.pg_size < 1 || pool_cfg.pg_size > 256 ||
+        pool_cfg.scheme !== 'replicated' && pool_cfg.pg_size < 3)
+    {
+        if (warn)
+            console.log('Pool '+pool_id+' has invalid pg_size');
+        return false;
+    }
+    if (!pool_cfg.pg_minsize || pool_cfg.pg_minsize < 1 || pool_cfg.pg_minsize > pool_cfg.pg_size ||
+        pool_cfg.scheme === 'xor' && pool_cfg.pg_minsize < (pool_cfg.pg_size - 1))
+    {
+        if (warn)
+            console.log('Pool '+pool_id+' has invalid pg_minsize');
+        return false;
+    }
+    if (pool_cfg.scheme === 'xor' && pool_cfg.parity_chunks != 0 && pool_cfg.parity_chunks != 1)
+    {
+        if (warn)
+            console.log('Pool '+pool_id+' has invalid parity_chunks (must be 1)');
+        return false;
+    }
+    if ((pool_cfg.scheme === 'ec' || pool_cfg.scheme === 'jerasure') &&
+        (pool_cfg.parity_chunks < 1 || pool_cfg.parity_chunks > pool_cfg.pg_size-2))
+    {
+        if (warn)
+            console.log('Pool '+pool_id+' has invalid parity_chunks (must be between 1 and pg_size-2)');
+        return false;
+    }
+    if (!pool_cfg.pg_count || pool_cfg.pg_count < 1)
+    {
+        if (warn)
+            console.log('Pool '+pool_id+' has invalid pg_count');
+        return false;
+    }
+    if (!pool_cfg.name)
+    {
+        if (warn)
+            console.log('Pool '+pool_id+' has empty name');
+        return false;
+    }
+    if (pool_cfg.max_osd_combinations < 100)
+    {
+        if (warn)
+            console.log('Pool '+pool_id+' has invalid max_osd_combinations (must be at least 100)');
+        return false;
+    }
+    if (pool_cfg.root_node && typeof(pool_cfg.root_node) != 'string')
+    {
+        if (warn)
+            console.log('Pool '+pool_id+' has invalid root_node (must be a string)');
+        return false;
+    }
+    if (pool_cfg.osd_tags && typeof(pool_cfg.osd_tags) != 'string' &&
+        (!(pool_cfg.osd_tags instanceof Array) || pool_cfg.osd_tags.filter(t => typeof t != 'string').length > 0))
+    {
+        if (warn)
+            console.log('Pool '+pool_id+' has invalid osd_tags (must be a string or array of strings)');
+        return false;
+    }
+    if (pool_cfg.primary_affinity_tags && typeof(pool_cfg.primary_affinity_tags) != 'string' &&
+        (!(pool_cfg.primary_affinity_tags instanceof Array) || pool_cfg.primary_affinity_tags.filter(t => typeof t != 'string').length > 0))
+    {
+        if (warn)
+            console.log('Pool '+pool_id+' has invalid primary_affinity_tags (must be a string or array of strings)');
+        return false;
+    }
+    if (!get_pg_rules(pool_id, pool_cfg, placement_levels, true))
+    {
+        return false;
+    }
+    return true;
+}
+
+function get_pg_rules(pool_id, pool_cfg, placement_levels, warn)
+{
+    if (pool_cfg.level_placement)
+    {
+        const pg_size = (0|pool_cfg.pg_size);
+        let rules = pool_cfg.level_placement;
+        if (typeof rules === 'string')
+        {
+            rules = rules.split(/\s+/).map(s => s.split(/=/, 2)).reduce((a, c) => { a[c[0]] = c[1]; return a; }, {});
+        }
+        else
+        {
+            rules = { ...rules };
+        }
+        // Always add failure_domain to prevent rules from being totally incorrect
+        const all_diff = [];
+        for (let i = 1; i <= pg_size; i++)
+        {
+            all_diff.push(i);
+        }
+        rules[pool_cfg.failure_domain || 'host'] = all_diff;
+        placement_levels = placement_levels||{};
+        placement_levels.host = placement_levels.host || 100;
+        placement_levels.osd = placement_levels.osd || 101;
+        for (const k in rules)
+        {
+            if (!placement_levels[k] || typeof rules[k] !== 'string' &&
+                (!(rules[k] instanceof Array) ||
+                rules[k].filter(s => typeof s !== 'string' && typeof s !== 'number').length > 0))
+            {
+                if (warn)
+                    console.log('Pool '+pool_id+' configuration is invalid: level_placement should be { [level]: string | (string|number)[] }');
+                return null;
+            }
+            else if (rules[k].length != pg_size)
+            {
+                if (warn)
+                    console.log('Pool '+pool_id+' configuration is invalid: values in level_placement should contain exactly pg_size ('+pg_size+') items');
+                return null;
+            }
+        }
+        return parse_level_indexes(rules);
+    }
+    else if (typeof pool_cfg.raw_placement === 'string')
+    {
+        try
+        {
+            return parse_pg_dsl(pool_cfg.raw_placement);
+        }
+        catch (e)
+        {
+            if (warn)
+                console.log('Pool '+pool_id+' configuration is invalid: invalid raw_placement: '+e.message);
+        }
+    }
+    else
+    {
+        let rules = [ [] ];
+        let prev = [ 1 ];
+        for (let i = 1; i < pool_cfg.pg_size; i++)
+        {
+            rules.push([ [ pool_cfg.failure_domain||'host', '!=', prev ] ]);
+            prev = [ ...prev, i+1 ];
+        }
+        return rules;
+    }
+}
+
+module.exports = {
+    validate_pool_cfg,
+    get_pg_rules,
+};

mon/stats.js

@@ -0,0 +1,286 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 (see README.md for details)
+
+function derive_osd_stats(st, prev, prev_diff)
+{
+    const diff = { op_stats: {}, subop_stats: {}, recovery_stats: {}, inode_stats: {} };
+    if (!st || !st.time || !prev || !prev.time || prev.time >= st.time)
+    {
+        return prev_diff || diff;
+    }
+    const timediff = BigInt(st.time*1000 - prev.time*1000);
+    for (const op in st.op_stats||{})
+    {
+        const pr = prev && prev.op_stats && prev.op_stats[op];
+        let c = st.op_stats[op];
+        c = { bytes: BigInt(c.bytes||0), usec: BigInt(c.usec||0), count: BigInt(c.count||0) };
+        const b = c.bytes - BigInt(pr && pr.bytes||0);
+        const us = c.usec - BigInt(pr && pr.usec||0);
+        const n = c.count - BigInt(pr && pr.count||0);
+        if (n > 0)
+            diff.op_stats[op] = { ...c, bps: b*1000n/timediff, iops: n*1000n/timediff, lat: us/n };
+    }
+    for (const op in st.subop_stats||{})
+    {
+        const pr = prev && prev.subop_stats && prev.subop_stats[op];
+        let c = st.subop_stats[op];
+        c = { usec: BigInt(c.usec||0), count: BigInt(c.count||0) };
+        const us = c.usec - BigInt(pr && pr.usec||0);
+        const n = c.count - BigInt(pr && pr.count||0);
+        if (n > 0)
+            diff.subop_stats[op] = { ...c, iops: n*1000n/timediff, lat: us/n };
+    }
+    for (const op in st.recovery_stats||{})
+    {
+        const pr = prev && prev.recovery_stats && prev.recovery_stats[op];
+        let c = st.recovery_stats[op];
+        c = { bytes: BigInt(c.bytes||0), count: BigInt(c.count||0) };
+        const b = c.bytes - BigInt(pr && pr.bytes||0);
+        const n = c.count - BigInt(pr && pr.count||0);
+        if (n > 0)
+            diff.recovery_stats[op] = { ...c, bps: b*1000n/timediff, iops: n*1000n/timediff };
+    }
+    for (const pool_id in st.inode_stats||{})
+    {
+        diff.inode_stats[pool_id] = {};
+        for (const inode_num in st.inode_stats[pool_id])
+        {
+            const inode_diff = diff.inode_stats[pool_id][inode_num] = {};
+            for (const op of [ 'read', 'write', 'delete' ])
+            {
+                const c = st.inode_stats[pool_id][inode_num][op];
+                const pr = prev && prev.inode_stats && prev.inode_stats[pool_id] &&
+                    prev.inode_stats[pool_id][inode_num] && prev.inode_stats[pool_id][inode_num][op];
+                const n = BigInt(c.count||0) - BigInt(pr && pr.count||0);
+                inode_diff[op] = {
+                    bps: (BigInt(c.bytes||0) - BigInt(pr && pr.bytes||0))*1000n/timediff,
+                    iops: n*1000n/timediff,
+                    lat: (BigInt(c.usec||0) - BigInt(pr && pr.usec||0))/(n || 1n),
+                };
+            }
+        }
+    }
+    return diff;
+}
+
+// sum_op_stats(this.state.osd, this.prev_stats)
+function sum_op_stats(all_osd, prev_stats)
+{
+    for (const osd in all_osd.stats)
+    {
+        const cur = { ...all_osd.stats[osd], inode_stats: all_osd.inodestats[osd]||{} };
+        prev_stats.osd_diff[osd] = derive_osd_stats(
+            cur, prev_stats.osd_stats[osd], prev_stats.osd_diff[osd]
+        );
+        prev_stats.osd_stats[osd] = cur;
+    }
+    const sum_diff = { op_stats: {}, subop_stats: {}, recovery_stats: {} };
+    // Sum derived values instead of deriving summed
+    for (const osd in all_osd.state)
+    {
+        const derived = prev_stats.osd_diff[osd];
+        if (!all_osd.state[osd] || !derived)
+        {
+            continue;
+        }
+        for (const type in sum_diff)
+        {
+            for (const op in derived[type]||{})
+            {
+                for (const k in derived[type][op])
+                {
+                    sum_diff[type][op] = sum_diff[type][op] || {};
+                    sum_diff[type][op][k] = (sum_diff[type][op][k] || 0n) + derived[type][op][k];
+                }
+            }
+        }
+    }
+    return sum_diff;
+}
+
+// sum_object_counts(this.state, this.config)
+function sum_object_counts(state, global_config)
+{
+    const object_counts = { object: 0n, clean: 0n, misplaced: 0n, degraded: 0n, incomplete: 0n };
+    const object_bytes = { object: 0n, clean: 0n, misplaced: 0n, degraded: 0n, incomplete: 0n };
+    for (const pool_id in state.pg.stats)
+    {
+        let object_size = 0;
+        for (const osd_num of state.pg.stats[pool_id].write_osd_set||[])
+        {
+            if (osd_num && state.osd.stats[osd_num] && state.osd.stats[osd_num].block_size)
+            {
+                object_size = state.osd.stats[osd_num].block_size;
+                break;
+            }
+        }
+        const pool_cfg = (state.config.pools[pool_id]||{});
+        if (!object_size)
+        {
+            object_size = pool_cfg.block_size || global_config.block_size || 131072;
+        }
+        if (pool_cfg.scheme !== 'replicated')
+        {
+            object_size *= ((pool_cfg.pg_size||0) - (pool_cfg.parity_chunks||0));
+        }
+        object_size = BigInt(object_size);
+        for (const pg_num in state.pg.stats[pool_id])
+        {
+            const st = state.pg.stats[pool_id][pg_num];
+            if (st)
+            {
+                for (const k in object_counts)
+                {
+                    if (st[k+'_count'])
+                    {
+                        object_counts[k] += BigInt(st[k+'_count']);
+                        object_bytes[k] += BigInt(st[k+'_count']) * object_size;
+                    }
+                }
+            }
+        }
+    }
+    return { object_counts, object_bytes };
+}
+
+// sum_inode_stats(this.state, this.prev_stats)
+function sum_inode_stats(state, prev_stats)
+{
+    const inode_stats = {};
+    const inode_stub = () => ({
+        raw_used: 0n,
+        read: { count: 0n, usec: 0n, bytes: 0n, bps: 0n, iops: 0n, lat: 0n },
+        write: { count: 0n, usec: 0n, bytes: 0n, bps: 0n, iops: 0n, lat: 0n },
+        delete: { count: 0n, usec: 0n, bytes: 0n, bps: 0n, iops: 0n, lat: 0n },
+    });
+    const seen_pools = {};
+    for (const pool_id in state.config.pools)
+    {
+        seen_pools[pool_id] = true;
+        state.pool.stats[pool_id] = state.pool.stats[pool_id] || {};
+        state.pool.stats[pool_id].used_raw_tb = 0n;
+    }
+    for (const osd_num in state.osd.space)
+    {
+        for (const pool_id in state.osd.space[osd_num])
+        {
+            state.pool.stats[pool_id] = state.pool.stats[pool_id] || {};
+            if (!seen_pools[pool_id])
+            {
+                state.pool.stats[pool_id].used_raw_tb = 0n;
+                seen_pools[pool_id] = true;
+            }
+            inode_stats[pool_id] = inode_stats[pool_id] || {};
+            for (const inode_num in state.osd.space[osd_num][pool_id])
+            {
+                const u = BigInt(state.osd.space[osd_num][pool_id][inode_num]||0);
+                if (inode_num)
+                {
+                    inode_stats[pool_id][inode_num] = inode_stats[pool_id][inode_num] || inode_stub();
+                    inode_stats[pool_id][inode_num].raw_used += u;
+                }
+                state.pool.stats[pool_id].used_raw_tb += u;
+            }
+        }
+    }
+    for (const pool_id in seen_pools)
+    {
+        const used = state.pool.stats[pool_id].used_raw_tb;
+        state.pool.stats[pool_id].used_raw_tb = Number(used)/1024/1024/1024/1024;
+    }
+    for (const osd_num in state.osd.state)
+    {
+        const ist = state.osd.inodestats[osd_num];
+        if (!ist || !state.osd.state[osd_num])
+        {
+            continue;
+        }
+        for (const pool_id in ist)
+        {
+            inode_stats[pool_id] = inode_stats[pool_id] || {};
+            for (const inode_num in ist[pool_id])
+            {
+                inode_stats[pool_id][inode_num] = inode_stats[pool_id][inode_num] || inode_stub();
+                for (const op of [ 'read', 'write', 'delete' ])
+                {
+                    inode_stats[pool_id][inode_num][op].count += BigInt(ist[pool_id][inode_num][op].count||0);
+                    inode_stats[pool_id][inode_num][op].usec += BigInt(ist[pool_id][inode_num][op].usec||0);
+                    inode_stats[pool_id][inode_num][op].bytes += BigInt(ist[pool_id][inode_num][op].bytes||0);
+                }
+            }
+        }
+    }
+    for (const osd in state.osd.state)
+    {
+        const osd_diff = prev_stats.osd_diff[osd];
+        if (!osd_diff || !state.osd.state[osd])
+        {
+            continue;
+        }
+        for (const pool_id in osd_diff.inode_stats)
+        {
+            for (const inode_num in prev_stats.osd_diff[osd].inode_stats[pool_id])
+            {
+                inode_stats[pool_id][inode_num] = inode_stats[pool_id][inode_num] || inode_stub();
+                for (const op of [ 'read', 'write', 'delete' ])
+                {
+                    const op_diff = prev_stats.osd_diff[osd].inode_stats[pool_id][inode_num][op] || {};
+                    const op_st = inode_stats[pool_id][inode_num][op];
+                    op_st.bps += op_diff.bps;
+                    op_st.iops += op_diff.iops;
+                    op_st.lat += op_diff.lat;
+                    op_st.n_osd = (op_st.n_osd || 0) + 1;
+                }
+            }
+        }
+    }
+    for (const pool_id in inode_stats)
+    {
+        for (const inode_num in inode_stats[pool_id])
+        {
+            let nonzero = inode_stats[pool_id][inode_num].raw_used > 0;
+            for (const op of [ 'read', 'write', 'delete' ])
+            {
+                const op_st = inode_stats[pool_id][inode_num][op];
+                if (op_st.n_osd)
+                {
+                    op_st.lat /= BigInt(op_st.n_osd);
+                    delete op_st.n_osd;
+                }
+                if (op_st.bps > 0 || op_st.iops > 0)
+                    nonzero = true;
+            }
+            if (!nonzero && (!state.config.inode[pool_id] || !state.config.inode[pool_id][inode_num]))
+            {
+                // Deleted inode (no data, no I/O, no config)
+                delete inode_stats[pool_id][inode_num];
+            }
+        }
+    }
+    return { inode_stats, seen_pools };
+}
+
+function serialize_bigints(obj)
+{
+    obj = { ...obj };
+    for (const k in obj)
+    {
+        if (typeof obj[k] == 'bigint')
+        {
+            obj[k] = ''+obj[k];
+        }
+        else if (typeof obj[k] == 'object')
+        {
+            obj[k] = serialize_bigints(obj[k]);
+        }
+    }
+    return obj;
+}
+
+module.exports = {
+    derive_osd_stats,
+    sum_op_stats,
+    sum_object_counts,
+    sum_inode_stats,
+    serialize_bigints,
+};

patches/qemu-make-patches.sh

@@ -9,7 +9,7 @@ for i in "$DIR"/qemu-*-vitastor.patch "$DIR"/pve-qemu-*-vitastor.patch; do
         echo '===================================================================' >> $i
         echo '--- /dev/null' >> $i
         echo '+++ a/block/vitastor.c' >> $i
-        echo '@@ -0,0 +1,'$(wc -l "$DIR"/../src/qemu_driver.c | cut -d ' ' -f 1)' @@' >> $i
+        echo '@@ -0,0 +1,'$(wc -l "$DIR"/../src/client/qemu_driver.c | cut -d ' ' -f 1)' @@' >> $i
-        cat "$DIR"/../src/qemu_driver.c | sed 's/^/+/' >> $i
+        cat "$DIR"/../src/client/qemu_driver.c | sed 's/^/+/' >> $i
     fi
 done

rpm/qemu-el8.Dockerfile

@@ -1,5 +1,5 @@
 # This is an attempt to automatically build patched RPM specs
-# More or less broken, better use *.spec.patch for now (and copy src/qemu_driver.c to SOURCES/qemu-vitastor.c)
+# More or less broken, better use *.spec.patch for now (and copy src/client/qemu_driver.c to SOURCES/qemu-vitastor.c)
 
 # Build packages for CentOS 8 inside a container
 # cd ..; podman build -t qemu-el8 -v `pwd`/packages:/root/packages -f rpm/qemu-el8.Dockerfile .

rpm/vitastor-el7.spec

@@ -108,10 +108,11 @@ npm install --production
 cd ..
 mkdir -p %buildroot/usr/lib/vitastor
 cp -r mon %buildroot/usr/lib/vitastor
+mv %buildroot/usr/lib/vitastor/mon/scripts/make-etcd %buildroot/usr/lib/vitastor/mon/
 mkdir -p %buildroot/lib/systemd/system
-cp mon/vitastor.target mon/vitastor-mon.service mon/vitastor-osd@.service %buildroot/lib/systemd/system
+cp mon/scripts/vitastor.target mon/scripts/vitastor-mon.service mon/scripts/vitastor-osd@.service %buildroot/lib/systemd/system
 mkdir -p %buildroot/lib/udev/rules.d
-cp mon/90-vitastor.rules %buildroot/lib/udev/rules.d
+cp mon/scripts/90-vitastor.rules %buildroot/lib/udev/rules.d
 
 
 %files

rpm/vitastor-el8.spec

@@ -105,10 +105,11 @@ npm install --production
 cd ..
 mkdir -p %buildroot/usr/lib/vitastor
 cp -r mon %buildroot/usr/lib/vitastor
+mv %buildroot/usr/lib/vitastor/mon/scripts/make-etcd %buildroot/usr/lib/vitastor/mon/
 mkdir -p %buildroot/lib/systemd/system
-cp mon/vitastor.target mon/vitastor-mon.service mon/vitastor-osd@.service %buildroot/lib/systemd/system
+cp mon/scripts/vitastor.target mon/scripts/vitastor-mon.service mon/scripts/vitastor-osd@.service %buildroot/lib/systemd/system
 mkdir -p %buildroot/lib/udev/rules.d
-cp mon/90-vitastor.rules %buildroot/lib/udev/rules.d
+cp mon/scripts/90-vitastor.rules %buildroot/lib/udev/rules.d
 
 
 %files

rpm/vitastor-el9.spec

@@ -98,10 +98,11 @@ npm install --production
 cd ..
 mkdir -p %buildroot/usr/lib/vitastor
 cp -r mon %buildroot/usr/lib/vitastor
+mv %buildroot/usr/lib/vitastor/mon/scripts/make-etcd %buildroot/usr/lib/vitastor/mon/
 mkdir -p %buildroot/lib/systemd/system
-cp mon/vitastor.target mon/vitastor-mon.service mon/vitastor-osd@.service %buildroot/lib/systemd/system
+cp mon/scripts/vitastor.target mon/scripts/vitastor-mon.service mon/scripts/vitastor-osd@.service %buildroot/lib/systemd/system
 mkdir -p %buildroot/lib/udev/rules.d
-cp mon/90-vitastor.rules %buildroot/lib/udev/rules.d
+cp mon/scripts/90-vitastor.rules %buildroot/lib/udev/rules.d
 
 
 %files

src/CMakeLists.txt

@@ -72,300 +72,28 @@ add_dependencies(test build_tests)
 
 include_directories(
 	../
+	${CMAKE_SOURCE_DIR}/src/blockstore
+	${CMAKE_SOURCE_DIR}/src/cmd
+	${CMAKE_SOURCE_DIR}/src/client
+	${CMAKE_SOURCE_DIR}/src/disk_tool
+	${CMAKE_SOURCE_DIR}/src/kv
+	${CMAKE_SOURCE_DIR}/src/nfs
+	${CMAKE_SOURCE_DIR}/src/osd
+	${CMAKE_SOURCE_DIR}/src/test
+	${CMAKE_SOURCE_DIR}/src/util
 	/usr/include/jerasure
 	${LIBURING_INCLUDE_DIRS}
 	${IBVERBS_INCLUDE_DIRS}
 )
 
-# libvitastor_blk.so
+add_subdirectory(blockstore)
-add_library(vitastor_blk SHARED
+add_subdirectory(cmd)
-	allocator.cpp blockstore.cpp blockstore_impl.cpp blockstore_disk.cpp blockstore_init.cpp blockstore_open.cpp blockstore_journal.cpp blockstore_read.cpp
+add_subdirectory(client)
-	blockstore_write.cpp blockstore_sync.cpp blockstore_stable.cpp blockstore_rollback.cpp blockstore_flush.cpp crc32c.c ringloop.cpp
+add_subdirectory(disk_tool)
-)
+add_subdirectory(kv)
-target_link_libraries(vitastor_blk
+add_subdirectory(nfs)
-	${LIBURING_LIBRARIES}
+add_subdirectory(osd)
-	tcmalloc_minimal
+add_subdirectory(test)
-	# for timerfd_manager
-	vitastor_common
-)
-set_target_properties(vitastor_blk PROPERTIES VERSION ${VERSION} SOVERSION 0)
-
-if (${WITH_FIO})
-	# libfio_vitastor_blk.so
-	add_library(fio_vitastor_blk SHARED
-		fio_engine.cpp
-		../json11/json11.cpp
-	)
-	target_link_libraries(fio_vitastor_blk
-		vitastor_blk
-	)
-endif (${WITH_FIO})
-
-# libvitastor_common.a
-set(MSGR_RDMA "")
-if (IBVERBS_LIBRARIES)
-	set(MSGR_RDMA "msgr_rdma.cpp")
-endif (IBVERBS_LIBRARIES)
-add_library(vitastor_common STATIC
-	epoll_manager.cpp etcd_state_client.cpp messenger.cpp addr_util.cpp
-	msgr_stop.cpp msgr_op.cpp msgr_send.cpp msgr_receive.cpp ringloop.cpp ../json11/json11.cpp
-	http_client.cpp osd_ops.cpp pg_states.cpp timerfd_manager.cpp str_util.cpp ${MSGR_RDMA}
-)
-target_compile_options(vitastor_common PUBLIC -fPIC)
-
-# vitastor-osd
-add_executable(vitastor-osd
-	osd_main.cpp osd.cpp osd_secondary.cpp osd_peering.cpp osd_flush.cpp osd_peering_pg.cpp
-	osd_primary.cpp osd_primary_chain.cpp osd_primary_sync.cpp osd_primary_write.cpp osd_primary_subops.cpp
-	osd_cluster.cpp osd_rmw.cpp osd_scrub.cpp osd_primary_describe.cpp
-)
-target_link_libraries(vitastor-osd
-	vitastor_common
-	vitastor_blk
-	Jerasure
-	${ISAL_LIBRARIES}
-	${IBVERBS_LIBRARIES}
-)
-
-if (${WITH_FIO})
-	# libfio_vitastor_sec.so
-	add_library(fio_vitastor_sec SHARED
-		fio_sec_osd.cpp
-		rw_blocking.cpp
-		addr_util.cpp
-	)
-	target_link_libraries(fio_vitastor_sec
-		tcmalloc_minimal
-	)
-endif (${WITH_FIO})
-
-# libvitastor_client.so
-add_library(vitastor_client SHARED
-	cluster_client.cpp
-	cluster_client_list.cpp
-	cluster_client_wb.cpp
-	vitastor_c.cpp
-	cli_common.cpp
-	cli_alloc_osd.cpp
-	cli_status.cpp
-	cli_describe.cpp
-	cli_fix.cpp
-	cli_ls.cpp
-	cli_create.cpp
-	cli_modify.cpp
-	cli_flatten.cpp
-	cli_merge.cpp
-	cli_rm_data.cpp
-	cli_rm.cpp
-	cli_rm_osd.cpp
-	cli_pool_cfg.cpp
-	cli_pool_create.cpp
-	cli_pool_ls.cpp
-	cli_pool_modify.cpp
-	cli_pool_rm.cpp
-)
-set_target_properties(vitastor_client PROPERTIES PUBLIC_HEADER "vitastor_c.h")
-target_link_libraries(vitastor_client
-	vitastor_common
-	${LIBURING_LIBRARIES}
-	${IBVERBS_LIBRARIES}
-)
-set_target_properties(vitastor_client PROPERTIES VERSION ${VERSION} SOVERSION 0)
-
-if (${WITH_FIO})
-	# libfio_vitastor.so
-	add_library(fio_vitastor SHARED
-		fio_cluster.cpp
-	)
-	target_link_libraries(fio_vitastor
-		vitastor_client
-	)
-endif (${WITH_FIO})
-
-# vitastor-nbd
-pkg_check_modules(NL3 libnl-3.0 libnl-genl-3.0)
-add_executable(vitastor-nbd
-	nbd_proxy.cpp
-)
-target_include_directories(vitastor-nbd PUBLIC ${NL3_INCLUDE_DIRS})
-target_link_libraries(vitastor-nbd vitastor_client ${NL3_LIBRARIES})
-if (HAVE_NBD_NETLINK_H AND NL3_LIBRARIES)
-	target_compile_definitions(vitastor-nbd PUBLIC HAVE_NBD_NETLINK_H)
-endif (HAVE_NBD_NETLINK_H AND NL3_LIBRARIES)
-
-# libvitastor_kv.so
-add_library(vitastor_kv SHARED
-	kv_db.cpp
-	kv_db.h
-)
-target_link_libraries(vitastor_kv
-	vitastor_client
-)
-set_target_properties(vitastor_kv PROPERTIES VERSION ${VERSION} SOVERSION 0)
-
-# vitastor-kv
-add_executable(vitastor-kv
-	kv_cli.cpp
-)
-target_link_libraries(vitastor-kv
-	vitastor_kv
-)
-
-add_executable(vitastor-kv-stress
-	kv_stress.cpp
-)
-target_link_libraries(vitastor-kv-stress
-	vitastor_kv
-)
-
-# vitastor-nfs
-add_executable(vitastor-nfs
-	nfs_proxy.cpp
-	nfs_block.cpp
-	nfs_kv.cpp
-	nfs_kv_create.cpp
-	nfs_kv_getattr.cpp
-	nfs_kv_link.cpp
-	nfs_kv_lookup.cpp
-	nfs_kv_read.cpp
-	nfs_kv_readdir.cpp
-	nfs_kv_remove.cpp
-	nfs_kv_rename.cpp
-	nfs_kv_setattr.cpp
-	nfs_kv_write.cpp
-	nfs_fsstat.cpp
-	nfs_mount.cpp
-	nfs_portmap.cpp
-	sha256.c
-	nfs/xdr_impl.cpp
-	nfs/rpc_xdr.cpp
-	nfs/portmap_xdr.cpp
-	nfs/nfs_xdr.cpp
-)
-target_link_libraries(vitastor-nfs
-	vitastor_client
-	vitastor_kv
-)
-
-# vitastor-cli
-add_executable(vitastor-cli
-	cli.cpp
-)
-target_link_libraries(vitastor-cli
-	vitastor_client
-)
-configure_file(vitastor.pc.in vitastor.pc @ONLY)
-
-# vitastor-disk
-add_executable(vitastor-disk
-	disk_tool.cpp disk_simple_offsets.cpp
-	disk_tool_journal.cpp disk_tool_meta.cpp disk_tool_prepare.cpp disk_tool_resize.cpp disk_tool_udev.cpp disk_tool_utils.cpp disk_tool_upgrade.cpp
-	crc32c.c str_util.cpp ../json11/json11.cpp rw_blocking.cpp allocator.cpp ringloop.cpp blockstore_disk.cpp
-)
-target_link_libraries(vitastor-disk
-	tcmalloc_minimal
-	${LIBURING_LIBRARIES}
-)
-
-if (${WITH_QEMU})
-	# qemu_driver.so
-	add_library(qemu_vitastor SHARED
-		qemu_driver.c
-	)
-	target_compile_options(qemu_vitastor PUBLIC -DVITASTOR_SOURCE_TREE)
-	target_include_directories(qemu_vitastor PUBLIC
-		../qemu/b/qemu
-		../qemu/include
-		${GLIB_INCLUDE_DIRS}
-	)
-	target_link_libraries(qemu_vitastor
-		vitastor_client
-	)
-	set_target_properties(qemu_vitastor PROPERTIES
-		PREFIX ""
-		OUTPUT_NAME "block-vitastor"
-	)
-endif (${WITH_QEMU})
-
-### Test stubs
-
-# stub_osd, stub_bench, osd_test
-add_executable(stub_osd stub_osd.cpp rw_blocking.cpp addr_util.cpp)
-target_link_libraries(stub_osd tcmalloc_minimal)
-add_executable(stub_bench stub_bench.cpp rw_blocking.cpp addr_util.cpp)
-target_link_libraries(stub_bench tcmalloc_minimal)
-add_executable(osd_test osd_test.cpp rw_blocking.cpp addr_util.cpp)
-target_link_libraries(osd_test tcmalloc_minimal)
-
-# osd_rmw_test
-add_executable(osd_rmw_test EXCLUDE_FROM_ALL osd_rmw_test.cpp allocator.cpp)
-target_link_libraries(osd_rmw_test Jerasure ${ISAL_LIBRARIES} tcmalloc_minimal)
-add_dependencies(build_tests osd_rmw_test)
-add_test(NAME osd_rmw_test COMMAND osd_rmw_test)
-
-if (ISAL_LIBRARIES)
-	add_executable(osd_rmw_test_je EXCLUDE_FROM_ALL osd_rmw_test.cpp allocator.cpp)
-	target_compile_definitions(osd_rmw_test_je PUBLIC -DNO_ISAL)
-	target_link_libraries(osd_rmw_test_je Jerasure tcmalloc_minimal)
-	add_dependencies(build_tests osd_rmw_test_je)
-	add_test(NAME osd_rmw_test_jerasure COMMAND osd_rmw_test_je)
-endif (ISAL_LIBRARIES)
-
-# stub_uring_osd
-add_executable(stub_uring_osd
-	stub_uring_osd.cpp
-)
-target_link_libraries(stub_uring_osd
-	vitastor_common
-	${LIBURING_LIBRARIES}
-	${IBVERBS_LIBRARIES}
-	tcmalloc_minimal
-)
-
-# osd_peering_pg_test
-add_executable(osd_peering_pg_test EXCLUDE_FROM_ALL osd_peering_pg_test.cpp osd_peering_pg.cpp)
-target_link_libraries(osd_peering_pg_test tcmalloc_minimal)
-add_dependencies(build_tests osd_peering_pg_test)
-add_test(NAME osd_peering_pg_test COMMAND osd_peering_pg_test)
-
-# test_allocator
-add_executable(test_allocator EXCLUDE_FROM_ALL test_allocator.cpp allocator.cpp)
-add_dependencies(build_tests test_allocator)
-add_test(NAME test_allocator COMMAND test_allocator)
-
-# test_cas
-add_executable(test_cas
-	test_cas.cpp
-)
-target_link_libraries(test_cas
-	vitastor_client
-)
-
-# test_crc32
-add_executable(test_crc32
-	test_crc32.cpp
-)
-target_link_libraries(test_crc32
-	vitastor_blk
-)
-
-# test_cluster_client
-add_executable(test_cluster_client
-	EXCLUDE_FROM_ALL
-	test_cluster_client.cpp
-	pg_states.cpp osd_ops.cpp cluster_client.cpp cluster_client_list.cpp cluster_client_wb.cpp msgr_op.cpp mock/messenger.cpp msgr_stop.cpp
-	etcd_state_client.cpp timerfd_manager.cpp str_util.cpp ../json11/json11.cpp
-)
-target_compile_definitions(test_cluster_client PUBLIC -D__MOCK__)
-target_include_directories(test_cluster_client PUBLIC ${CMAKE_SOURCE_DIR}/src/mock)
-add_dependencies(build_tests test_cluster_client)
-add_test(NAME test_cluster_client COMMAND test_cluster_client)
-
-## test_blockstore, test_shit
-#add_executable(test_blockstore test_blockstore.cpp)
-#target_link_libraries(test_blockstore blockstore)
-#add_executable(test_shit test_shit.cpp osd_peering_pg.cpp)
-#target_link_libraries(test_shit ${LIBURING_LIBRARIES} m)
 
 ### Install
 
@@ -378,7 +106,7 @@ install(
 	LIBRARY DESTINATION ${CMAKE_INSTALL_LIBDIR}
 	PUBLIC_HEADER DESTINATION ${CMAKE_INSTALL_INCLUDEDIR}
 )
-install(FILES ${CMAKE_CURRENT_BINARY_DIR}/vitastor.pc DESTINATION ${CMAKE_INSTALL_LIBDIR}/pkgconfig)
+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/client/vitastor.pc DESTINATION ${CMAKE_INSTALL_LIBDIR}/pkgconfig)
 if (${WITH_FIO})
 	install(TARGETS fio_vitastor fio_vitastor_blk fio_vitastor_sec LIBRARY DESTINATION ${CMAKE_INSTALL_LIBDIR})
 endif (${WITH_FIO})

src/blockstore/CMakeLists.txt

@@ -0,0 +1,27 @@
+cmake_minimum_required(VERSION 2.8.12)
+
+project(vitastor)
+
+# libvitastor_blk.so
+add_library(vitastor_blk SHARED
+	../util/allocator.cpp blockstore.cpp blockstore_impl.cpp blockstore_disk.cpp blockstore_init.cpp blockstore_open.cpp blockstore_journal.cpp blockstore_read.cpp
+	blockstore_write.cpp blockstore_sync.cpp blockstore_stable.cpp blockstore_rollback.cpp blockstore_flush.cpp ../util/crc32c.c ../util/ringloop.cpp
+)
+target_link_libraries(vitastor_blk
+	${LIBURING_LIBRARIES}
+	tcmalloc_minimal
+	# for timerfd_manager
+	vitastor_common
+)
+set_target_properties(vitastor_blk PROPERTIES VERSION ${VERSION} SOVERSION 0)
+
+if (${WITH_FIO})
+	# libfio_vitastor_blk.so
+	add_library(fio_vitastor_blk SHARED
+		fio_engine.cpp
+		../../json11/json11.cpp
+	)
+	target_link_libraries(fio_vitastor_blk
+		vitastor_blk
+	)
+endif (${WITH_FIO})

src/client/CMakeLists.txt

@@ -0,0 +1,95 @@
+cmake_minimum_required(VERSION 2.8.12)
+
+project(vitastor)
+
+# libvitastor_common.a
+set(MSGR_RDMA "")
+if (IBVERBS_LIBRARIES)
+	set(MSGR_RDMA "msgr_rdma.cpp")
+endif (IBVERBS_LIBRARIES)
+add_library(vitastor_common STATIC
+	../util/epoll_manager.cpp etcd_state_client.cpp messenger.cpp ../util/addr_util.cpp
+	msgr_stop.cpp msgr_op.cpp msgr_send.cpp msgr_receive.cpp ../util/ringloop.cpp ../../json11/json11.cpp
+	http_client.cpp osd_ops.cpp pg_states.cpp ../util/timerfd_manager.cpp ../util/str_util.cpp ${MSGR_RDMA}
+)
+target_compile_options(vitastor_common PUBLIC -fPIC)
+
+# libvitastor_client.so
+add_library(vitastor_client SHARED
+	cluster_client.cpp
+	cluster_client_list.cpp
+	cluster_client_wb.cpp
+	vitastor_c.cpp
+)
+set_target_properties(vitastor_client PROPERTIES PUBLIC_HEADER "client/vitastor_c.h")
+target_link_libraries(vitastor_client
+	vitastor_common
+	vitastor_cli
+	${LIBURING_LIBRARIES}
+	${IBVERBS_LIBRARIES}
+)
+set_target_properties(vitastor_client PROPERTIES VERSION ${VERSION} SOVERSION 0)
+configure_file(vitastor.pc.in vitastor.pc @ONLY)
+
+if (${WITH_FIO})
+	# libfio_vitastor.so
+	add_library(fio_vitastor SHARED
+		fio_cluster.cpp
+	)
+	target_link_libraries(fio_vitastor
+		vitastor_client
+	)
+
+	# libfio_vitastor_sec.so
+	add_library(fio_vitastor_sec SHARED
+		fio_sec_osd.cpp
+		../util/rw_blocking.cpp
+		../util/addr_util.cpp
+	)
+	target_link_libraries(fio_vitastor_sec
+		tcmalloc_minimal
+	)
+endif (${WITH_FIO})
+
+# vitastor-nbd
+pkg_check_modules(NL3 libnl-3.0 libnl-genl-3.0)
+add_executable(vitastor-nbd
+	nbd_proxy.cpp
+)
+target_include_directories(vitastor-nbd PUBLIC ${NL3_INCLUDE_DIRS})
+target_link_libraries(vitastor-nbd vitastor_client ${NL3_LIBRARIES})
+if (HAVE_NBD_NETLINK_H AND NL3_LIBRARIES)
+	target_compile_definitions(vitastor-nbd PUBLIC HAVE_NBD_NETLINK_H)
+endif (HAVE_NBD_NETLINK_H AND NL3_LIBRARIES)
+
+if (${WITH_QEMU})
+	# qemu_driver.so
+	add_library(qemu_vitastor SHARED
+		qemu_driver.c
+	)
+	target_compile_options(qemu_vitastor PUBLIC -DVITASTOR_SOURCE_TREE)
+	target_include_directories(qemu_vitastor PUBLIC
+		../../qemu/b/qemu
+		../../qemu/include
+		${GLIB_INCLUDE_DIRS}
+	)
+	target_link_libraries(qemu_vitastor
+		vitastor_client
+	)
+	set_target_properties(qemu_vitastor PROPERTIES
+		PREFIX ""
+		OUTPUT_NAME "block-vitastor"
+	)
+endif (${WITH_QEMU})
+
+# test_cluster_client
+add_executable(test_cluster_client
+	EXCLUDE_FROM_ALL
+	../test/test_cluster_client.cpp
+	pg_states.cpp osd_ops.cpp cluster_client.cpp cluster_client_list.cpp cluster_client_wb.cpp msgr_op.cpp ../test/mock/messenger.cpp msgr_stop.cpp
+	etcd_state_client.cpp ../util/timerfd_manager.cpp ../util/str_util.cpp ../../json11/json11.cpp
+)
+target_compile_definitions(test_cluster_client PUBLIC -D__MOCK__)
+target_include_directories(test_cluster_client BEFORE PUBLIC ${CMAKE_SOURCE_DIR}/src/test/mock)
+add_dependencies(build_tests test_cluster_client)
+add_test(NAME test_cluster_client COMMAND test_cluster_client)

src/client/etcd_state_client.h

@@ -20,9 +20,11 @@
 #define MAX_DATA_BLOCK_SIZE 128*1024*1024
 #define DEFAULT_BITMAP_GRANULARITY 4096
 
+#ifndef IMMEDIATE_NONE
 #define IMMEDIATE_NONE 0
 #define IMMEDIATE_SMALL 1
 #define IMMEDIATE_ALL 2
+#endif
 
 struct etcd_kv_t
 {

src/client/vitastor_c.cpp

@@ -200,6 +200,11 @@ vitastor_c *vitastor_c_create_epoll_json(const char **options, int options_len)
     return self;
 }
 
+void* vitastor_c_get_internal_client(vitastor_c *client)
+{
+    return client->cli;
+}
+
 void vitastor_c_destroy(vitastor_c *client)
 {
     delete client->cli;
@@ -379,4 +384,12 @@ int vitastor_c_inode_get_readonly(void *handle)
     return watch->cfg.readonly;
 }
 
+uint32_t vitastor_c_inode_get_immediate_commit(vitastor_c *client, uint64_t inode_num)
+{
+    auto pool_it = client->cli->st_cli.pool_config.find(INODE_POOL(inode_num));
+    if (pool_it == client->cli->st_cli.pool_config.end())
+        return 0;
+    return pool_it->second.immediate_commit;
+}
+
 }

src/client/vitastor_c.h

@@ -7,11 +7,18 @@
 #define VITASTOR_QEMU_PROXY_H
 
 // C API wrapper version
-#define VITASTOR_C_API_VERSION 3
+#define VITASTOR_C_API_VERSION 4
 
 #ifndef POOL_ID_BITS
 #define POOL_ID_BITS 16
 #endif
+
+#ifndef IMMEDIATE_NONE
+#define IMMEDIATE_NONE 0
+#define IMMEDIATE_SMALL 1
+#define IMMEDIATE_ALL 2
+#endif
+
 #include <stdint.h>
 #include <sys/uio.h>
 
@@ -41,6 +48,7 @@ vitastor_c *vitastor_c_create_uring(const char *config_path, const char *etcd_ho
     int use_rdma, const char *rdma_device, int rdma_port_num, int rdma_gid_index, int rdma_mtu, int log_level);
 vitastor_c *vitastor_c_create_uring_json(const char **options, int options_len);
 vitastor_c *vitastor_c_create_epoll_json(const char **options, int options_len);
+void* vitastor_c_get_internal_client(vitastor_c *client);
 void vitastor_c_destroy(vitastor_c *client);
 int vitastor_c_is_ready(vitastor_c *client);
 int vitastor_c_uring_register_eventfd(vitastor_c *client);
@@ -64,6 +72,8 @@ uint64_t vitastor_c_inode_get_num(void *handle);
 uint32_t vitastor_c_inode_get_block_size(vitastor_c *client, uint64_t inode_num);
 uint32_t vitastor_c_inode_get_bitmap_granularity(vitastor_c *client, uint64_t inode_num);
 int vitastor_c_inode_get_readonly(void *handle);
+// Returns IMMEDIATE_ALL, IMMEDIATE_SMALL or IMMEDIATE_NONE
+uint32_t vitastor_c_inode_get_immediate_commit(vitastor_c *client, uint64_t inode_num);
 
 #ifdef __cplusplus
 }

src/cmd/CMakeLists.txt

@@ -0,0 +1,34 @@
+cmake_minimum_required(VERSION 2.8.12)
+
+project(vitastor)
+
+# libvitastor_cli.a
+add_library(vitastor_cli STATIC
+	cli_common.cpp
+	cli_alloc_osd.cpp
+	cli_status.cpp
+	cli_describe.cpp
+	cli_fix.cpp
+	cli_ls.cpp
+	cli_create.cpp
+	cli_modify.cpp
+	cli_flatten.cpp
+	cli_merge.cpp
+	cli_rm_data.cpp
+	cli_rm.cpp
+	cli_rm_osd.cpp
+	cli_pool_cfg.cpp
+	cli_pool_create.cpp
+	cli_pool_ls.cpp
+	cli_pool_modify.cpp
+	cli_pool_rm.cpp
+)
+target_compile_options(vitastor_cli PUBLIC -fPIC)
+
+# vitastor-cli
+add_executable(vitastor-cli
+	cli.cpp
+)
+target_link_libraries(vitastor-cli
+	vitastor_client
+)