2020-09-17 23:02:40 +03:00
|
|
|
// Copyright (c) Vitaliy Filippov, 2019+
|
2021-02-06 01:26:07 +03:00
|
|
|
// License: VNPL-1.1 (see README.md for details)
|
2020-09-17 23:02:40 +03:00
|
|
|
|
2021-11-28 01:19:42 +03:00
|
|
|
const fs = require('fs');
|
2020-05-12 17:49:18 +03:00
|
|
|
const http = require('http');
|
2020-09-01 18:50:23 +03:00
|
|
|
const crypto = require('crypto');
|
2020-05-12 17:49:18 +03:00
|
|
|
const os = require('os');
|
|
|
|
const WebSocket = require('ws');
|
|
|
|
const LPOptimizer = require('./lp-optimizer.js');
|
|
|
|
const stableStringify = require('./stable-stringify.js');
|
2020-09-01 02:22:50 +03:00
|
|
|
const PGUtil = require('./PGUtil.js');
|
2020-05-12 17:49:18 +03:00
|
|
|
|
2020-11-08 01:54:12 +03:00
|
|
|
// FIXME document all etcd keys and config variables in the form of JSON schema or similar
|
2020-03-12 00:56:41 +03:00
|
|
|
const etcd_nonempty_keys = {
|
|
|
|
'config/global': 1,
|
|
|
|
'config/node_placement': 1,
|
|
|
|
'config/pools': 1,
|
|
|
|
'config/pgs': 1,
|
|
|
|
'history/last_clean_pgs': 1,
|
|
|
|
'stats': 1,
|
|
|
|
};
|
2020-11-08 01:54:12 +03:00
|
|
|
const etcd_allow = new RegExp('^'+[
|
|
|
|
'config/global',
|
|
|
|
'config/node_placement',
|
|
|
|
'config/pools',
|
|
|
|
'config/osd/[1-9]\\d*',
|
|
|
|
'config/pgs',
|
2021-01-30 01:35:58 +03:00
|
|
|
'config/inode/[1-9]\\d*/[1-9]\\d*',
|
2020-11-08 01:54:12 +03:00
|
|
|
'osd/state/[1-9]\\d*',
|
|
|
|
'osd/stats/[1-9]\\d*',
|
2021-01-21 00:30:18 +03:00
|
|
|
'osd/inodestats/[1-9]\\d*',
|
2021-01-19 02:14:13 +03:00
|
|
|
'osd/space/[1-9]\\d*',
|
2020-11-08 01:54:12 +03:00
|
|
|
'mon/master',
|
|
|
|
'pg/state/[1-9]\\d*/[1-9]\\d*',
|
|
|
|
'pg/stats/[1-9]\\d*/[1-9]\\d*',
|
|
|
|
'pg/history/[1-9]\\d*/[1-9]\\d*',
|
2021-11-06 13:38:52 +03:00
|
|
|
'pool/stats/[1-9]\\d*',
|
2021-03-06 01:30:42 +03:00
|
|
|
'history/last_clean_pgs',
|
2021-04-04 11:52:11 +03:00
|
|
|
'inode/stats/[1-9]\\d*/[1-9]\\d*',
|
2021-10-21 14:00:54 +03:00
|
|
|
'pool/stats/[1-9]\\d*',
|
2020-11-08 01:54:12 +03:00
|
|
|
'stats',
|
2021-07-02 22:47:01 +03:00
|
|
|
'index/image/.*',
|
|
|
|
'index/maxid/[1-9]\\d*',
|
2020-11-08 01:54:12 +03:00
|
|
|
].join('$|^')+'$');
|
2020-09-01 17:07:06 +03:00
|
|
|
|
2020-11-08 01:54:12 +03:00
|
|
|
const etcd_tree = {
|
|
|
|
config: {
|
|
|
|
/* global: {
|
2021-04-30 01:21:16 +03:00
|
|
|
// WARNING: NOT ALL OF THESE ARE ACTUALLY CONFIGURABLE HERE
|
2021-06-20 00:23:56 +03:00
|
|
|
// THIS IS JUST A POOR MAN'S CONFIG DOCUMENTATION
|
2021-04-30 01:21:16 +03:00
|
|
|
// etcd connection
|
|
|
|
config_path: "/etc/vitastor/vitastor.conf",
|
|
|
|
etcd_address: "10.0.115.10:2379/v3",
|
|
|
|
etcd_prefix: "/vitastor",
|
2020-11-08 01:54:12 +03:00
|
|
|
// mon
|
|
|
|
etcd_mon_ttl: 30, // min: 10
|
|
|
|
etcd_mon_timeout: 1000, // ms. min: 0
|
|
|
|
etcd_mon_retries: 5, // min: 0
|
|
|
|
mon_change_timeout: 1000, // ms. min: 100
|
|
|
|
mon_stats_timeout: 1000, // ms. min: 100
|
2021-03-09 01:24:51 +03:00
|
|
|
osd_out_time: 600, // seconds. min: 0
|
2020-11-08 01:54:12 +03:00
|
|
|
placement_levels: { datacenter: 1, rack: 2, host: 3, osd: 4, ... },
|
|
|
|
// client and osd
|
2021-04-30 01:21:16 +03:00
|
|
|
tcp_header_buffer_size: 65536,
|
2020-11-08 01:54:12 +03:00
|
|
|
use_sync_send_recv: false,
|
2021-04-30 01:21:16 +03:00
|
|
|
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: 32,
|
|
|
|
rdma_max_recv: 8,
|
|
|
|
rdma_max_msg: 1048576,
|
2020-11-08 01:54:12 +03:00
|
|
|
log_level: 0,
|
|
|
|
block_size: 131072,
|
|
|
|
disk_alignment: 4096,
|
|
|
|
bitmap_granularity: 4096,
|
|
|
|
immediate_commit: false, // 'all' or 'small'
|
|
|
|
client_dirty_limit: 33554432,
|
|
|
|
peer_connect_interval: 5, // seconds. min: 1
|
|
|
|
peer_connect_timeout: 5, // seconds. min: 1
|
2021-03-11 00:36:23 +03:00
|
|
|
osd_idle_timeout: 5, // seconds. min: 1
|
|
|
|
osd_ping_timeout: 5, // seconds. min: 1
|
2020-11-08 01:54:12 +03:00
|
|
|
up_wait_retry_interval: 500, // ms. min: 50
|
|
|
|
// osd
|
2021-11-09 01:27:12 +03:00
|
|
|
etcd_report_interval: 5,
|
2020-11-08 01:54:12 +03:00
|
|
|
run_primary: true,
|
2021-11-23 21:59:22 +03:00
|
|
|
osd_network: null, // "192.168.7.0/24" or an array of masks
|
2020-11-08 01:54:12 +03:00
|
|
|
bind_address: "0.0.0.0",
|
|
|
|
bind_port: 0,
|
|
|
|
autosync_interval: 5,
|
2021-10-30 14:26:48 +03:00
|
|
|
autosync_writes: 128,
|
2020-11-08 01:54:12 +03:00
|
|
|
client_queue_depth: 128, // unused
|
|
|
|
recovery_queue_depth: 4,
|
2021-03-15 02:26:39 +03:00
|
|
|
recovery_sync_batch: 16,
|
2020-11-08 01:54:12 +03:00
|
|
|
readonly: false,
|
2021-03-11 00:36:23 +03:00
|
|
|
no_recovery: false,
|
|
|
|
no_rebalance: false,
|
2020-11-08 01:54:12 +03:00
|
|
|
print_stats_interval: 3,
|
2021-03-11 00:36:23 +03:00
|
|
|
slow_log_interval: 10,
|
2020-11-08 01:54:12 +03:00
|
|
|
// 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
|
2021-03-11 00:36:23 +03:00
|
|
|
max_write_iodepth,
|
2021-03-24 01:58:33 +03:00
|
|
|
min_flusher_count: 1,
|
|
|
|
max_flusher_count: 256,
|
2020-11-08 01:54:12 +03:00
|
|
|
inmemory_metadata,
|
|
|
|
inmemory_journal,
|
|
|
|
journal_sector_buffer_count,
|
|
|
|
journal_no_same_sector_overwrites,
|
|
|
|
}, */
|
|
|
|
global: {},
|
|
|
|
/* node_placement: {
|
|
|
|
host1: { level: 'host', parent: 'rack1' },
|
|
|
|
...
|
|
|
|
}, */
|
|
|
|
node_placement: {},
|
|
|
|
/* pools: {
|
|
|
|
<id>: {
|
|
|
|
name: 'testpool',
|
2020-11-30 00:08:25 +03:00
|
|
|
// jerasure uses Reed-Solomon-Vandermonde codes
|
|
|
|
scheme: 'replicated' | 'xor' | 'jerasure',
|
2020-11-08 01:54:12 +03:00
|
|
|
pg_size: 3,
|
|
|
|
pg_minsize: 2,
|
2020-11-30 00:08:25 +03:00
|
|
|
// number of parity chunks, required for jerasure
|
|
|
|
parity_chunks?: 1,
|
2020-11-08 01:54:12 +03:00
|
|
|
pg_count: 100,
|
|
|
|
failure_domain: 'host',
|
|
|
|
max_osd_combinations: 10000,
|
|
|
|
pg_stripe_size: 4194304,
|
|
|
|
root_node?: 'rack1',
|
|
|
|
// restrict pool to OSDs having all of these tags
|
|
|
|
osd_tags?: 'nvme' | [ 'nvme', ... ],
|
2020-09-01 02:13:03 +03:00
|
|
|
},
|
2020-11-08 01:54:12 +03:00
|
|
|
...
|
|
|
|
}, */
|
|
|
|
pools: {},
|
|
|
|
osd: {
|
|
|
|
/* <id>: { reweight?: 1, tags?: [ 'nvme', ... ] }, ... */
|
|
|
|
},
|
|
|
|
/* pgs: {
|
|
|
|
hash: string,
|
|
|
|
items: {
|
|
|
|
<pool_id>: {
|
|
|
|
<pg_id>: {
|
|
|
|
osd_set: [ 1, 2, 3 ],
|
|
|
|
primary: 1,
|
|
|
|
pause: false,
|
2020-09-01 16:29:42 +03:00
|
|
|
}
|
|
|
|
}
|
2020-11-08 01:54:12 +03:00
|
|
|
}
|
|
|
|
}, */
|
|
|
|
pgs: {},
|
2021-01-30 01:35:58 +03:00
|
|
|
/* inode: {
|
|
|
|
<pool_id>: {
|
|
|
|
<inode_t>: {
|
|
|
|
name: string,
|
2021-02-10 01:21:57 +03:00
|
|
|
size?: uint64_t, // bytes
|
2021-01-30 01:35:58 +03:00
|
|
|
parent_pool?: <pool_id>,
|
|
|
|
parent_id?: <inode_t>,
|
|
|
|
readonly?: boolean,
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}, */
|
|
|
|
inode: {},
|
2020-11-08 01:54:12 +03:00
|
|
|
},
|
|
|
|
osd: {
|
|
|
|
state: {
|
|
|
|
/* <osd_num_t>: {
|
|
|
|
state: "up",
|
|
|
|
addresses: string[],
|
|
|
|
host: string,
|
|
|
|
port: uint16_t,
|
|
|
|
primary_enabled: boolean,
|
|
|
|
blockstore_enabled: boolean,
|
2020-09-01 16:29:42 +03:00
|
|
|
}, */
|
2020-05-12 17:49:18 +03:00
|
|
|
},
|
2020-11-08 01:54:12 +03:00
|
|
|
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 },
|
|
|
|
},
|
|
|
|
}, */
|
2020-05-12 17:49:18 +03:00
|
|
|
},
|
2021-01-21 00:30:18 +03:00
|
|
|
inodestats: {
|
|
|
|
/* <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 },
|
|
|
|
}, */
|
|
|
|
},
|
2021-01-19 02:14:13 +03:00
|
|
|
space: {
|
|
|
|
/* <osd_num_t>: {
|
|
|
|
<inode_t>: uint64_t, // bytes
|
|
|
|
}, */
|
|
|
|
},
|
2020-11-08 01:54:12 +03:00
|
|
|
},
|
|
|
|
mon: {
|
|
|
|
master: {
|
|
|
|
/* ip: [ string ], */
|
2020-05-12 17:49:18 +03:00
|
|
|
},
|
2020-11-08 01:54:12 +03:00
|
|
|
},
|
|
|
|
pg: {
|
|
|
|
state: {
|
|
|
|
/* <pool_id>: {
|
|
|
|
<pg_id>: {
|
|
|
|
primary: osd_num_t,
|
2021-11-13 02:06:54 +03:00
|
|
|
state: ("starting"|"peering"|"peered"|"incomplete"|"active"|"repeering"|"stopping"|"offline"|
|
2020-11-08 01:54:12 +03:00
|
|
|
"degraded"|"has_incomplete"|"has_degraded"|"has_misplaced"|"has_unclean"|
|
|
|
|
"has_invalid"|"left_on_dead")[],
|
|
|
|
}
|
|
|
|
}, */
|
2020-05-12 17:49:18 +03:00
|
|
|
},
|
2020-09-01 18:50:23 +03:00
|
|
|
stats: {
|
2020-11-08 01:54:12 +03:00
|
|
|
/* <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[],
|
|
|
|
},
|
2020-09-01 18:50:23 +03:00
|
|
|
}, */
|
|
|
|
},
|
2020-11-08 01:54:12 +03:00
|
|
|
history: {
|
|
|
|
/* <pool_id>: {
|
|
|
|
<pg_id>: {
|
|
|
|
osd_sets: osd_num_t[][],
|
|
|
|
all_peers: osd_num_t[],
|
|
|
|
epoch: uint32_t,
|
|
|
|
},
|
|
|
|
}, */
|
|
|
|
},
|
|
|
|
},
|
2021-01-19 02:14:13 +03:00
|
|
|
inode: {
|
2021-01-21 00:30:18 +03:00
|
|
|
stats: {
|
2021-06-20 00:23:56 +03:00
|
|
|
/* <pool_id>: {
|
|
|
|
<inode_t>: {
|
|
|
|
raw_used: uint64_t, // raw used bytes on OSDs
|
2021-11-10 23:51:15 +03:00
|
|
|
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 },
|
2021-06-20 00:23:56 +03:00
|
|
|
},
|
2021-01-19 02:14:13 +03:00
|
|
|
}, */
|
|
|
|
},
|
|
|
|
},
|
2021-07-02 22:47:01 +03:00
|
|
|
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
|
|
|
|
} */
|
|
|
|
},
|
|
|
|
},
|
2020-11-08 01:54:12 +03:00
|
|
|
stats: {
|
|
|
|
/* op_stats: {
|
2021-11-10 23:51:15 +03:00
|
|
|
<string>: { count: uint64_t, usec: uint64_t, bytes: uint64_t, bps: uint64_t, iops: uint64_t, lat: uint64_t },
|
2020-11-08 01:54:12 +03:00
|
|
|
},
|
|
|
|
subop_stats: {
|
2021-11-10 23:51:15 +03:00
|
|
|
<string>: { count: uint64_t, usec: uint64_t, iops: uint64_t, lat: uint64_t },
|
2020-11-08 01:54:12 +03:00
|
|
|
},
|
|
|
|
recovery_stats: {
|
2021-11-10 23:51:15 +03:00
|
|
|
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 },
|
2020-11-08 01:54:12 +03:00
|
|
|
},
|
|
|
|
object_counts: {
|
|
|
|
object: uint64_t,
|
|
|
|
clean: uint64_t,
|
|
|
|
misplaced: uint64_t,
|
|
|
|
degraded: uint64_t,
|
|
|
|
incomplete: uint64_t,
|
|
|
|
}, */
|
|
|
|
},
|
2021-03-06 01:30:42 +03:00
|
|
|
history: {
|
|
|
|
last_clean_pgs: {},
|
|
|
|
},
|
2021-07-02 22:47:01 +03:00
|
|
|
index: {
|
|
|
|
image: {
|
|
|
|
/* <name>: {
|
|
|
|
id: uint64_t,
|
|
|
|
pool_id: uint64_t,
|
|
|
|
}, */
|
|
|
|
},
|
|
|
|
maxid: {
|
|
|
|
/* <pool_id>: uint64_t, */
|
|
|
|
},
|
|
|
|
},
|
2020-11-08 01:54:12 +03:00
|
|
|
};
|
2020-05-12 17:49:18 +03:00
|
|
|
|
2020-11-08 01:54:12 +03:00
|
|
|
// FIXME Split into several files
|
|
|
|
class Mon
|
|
|
|
{
|
2020-05-12 17:49:18 +03:00
|
|
|
constructor(config)
|
|
|
|
{
|
2021-11-28 01:19:42 +03:00
|
|
|
this.die = (e) => this._die(e);
|
|
|
|
if (fs.existsSync(config.config_path||'/etc/vitastor/vitastor.conf'))
|
2020-05-12 17:49:18 +03:00
|
|
|
{
|
2021-11-28 01:19:42 +03:00
|
|
|
config = {
|
|
|
|
...JSON.parse(fs.readFileSync(config.config_path||'/etc/vitastor/vitastor.conf', { encoding: 'utf-8' })),
|
|
|
|
...config,
|
|
|
|
};
|
2020-05-12 17:49:18 +03:00
|
|
|
}
|
2021-11-28 01:19:42 +03:00
|
|
|
this.parse_etcd_addresses(config.etcd_address||config.etcd_url);
|
2020-09-11 18:49:03 +03:00
|
|
|
this.verbose = config.verbose || 0;
|
2021-11-28 01:19:42 +03:00
|
|
|
this.initConfig = config;
|
2020-09-11 19:25:52 +03:00
|
|
|
this.config = {};
|
2020-08-03 23:50:50 +03:00
|
|
|
this.etcd_prefix = config.etcd_prefix || '/vitastor';
|
2020-05-14 19:29:35 +03:00
|
|
|
this.etcd_prefix = this.etcd_prefix.replace(/\/\/+/g, '/').replace(/^\/?(.*[^\/])\/?$/, '/$1');
|
2020-05-12 17:49:18 +03:00
|
|
|
this.etcd_start_timeout = (config.etcd_start_timeout || 5) * 1000;
|
2020-09-01 17:07:06 +03:00
|
|
|
this.state = JSON.parse(JSON.stringify(this.constructor.etcd_tree));
|
2021-11-06 13:38:52 +03:00
|
|
|
this.signals_set = false;
|
2021-11-28 01:43:31 +03:00
|
|
|
this.on_stop_cb = () => this.on_stop(0).catch(console.error);
|
2020-05-12 17:49:18 +03:00
|
|
|
}
|
|
|
|
|
2021-11-28 01:19:42 +03:00
|
|
|
parse_etcd_addresses(addrs)
|
|
|
|
{
|
|
|
|
const is_local_ip = this.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);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2020-05-12 17:49:18 +03:00
|
|
|
async start()
|
|
|
|
{
|
|
|
|
await this.load_config();
|
|
|
|
await this.get_lease();
|
|
|
|
await this.become_master();
|
|
|
|
await this.load_cluster_state();
|
2020-09-05 02:14:43 +03:00
|
|
|
await this.start_watcher(this.config.etcd_mon_retries);
|
2021-11-06 13:38:52 +03:00
|
|
|
for (const pool_id in this.state.config.pools)
|
|
|
|
{
|
|
|
|
if (!this.state.pool.stats[pool_id] ||
|
|
|
|
!this.state.pool.stats[pool_id].pg_real_size)
|
|
|
|
{
|
|
|
|
// Generate missing data in etcd
|
|
|
|
this.state.config.pgs.hash = null;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
2020-05-12 17:49:18 +03:00
|
|
|
await this.recheck_pgs();
|
2021-11-09 01:41:22 +03:00
|
|
|
this.schedule_update_stats();
|
2020-05-12 17:49:18 +03:00
|
|
|
}
|
|
|
|
|
|
|
|
async load_config()
|
|
|
|
{
|
2020-09-05 02:14:43 +03:00
|
|
|
const res = await this.etcd_call('/kv/txn', { success: [
|
2020-05-12 17:49:18 +03:00
|
|
|
{ requestRange: { key: b64(this.etcd_prefix+'/config/global') } }
|
|
|
|
] }, this.etcd_start_timeout, -1);
|
2021-01-31 22:49:36 +03:00
|
|
|
if (res.responses[0].response_range.kvs)
|
|
|
|
{
|
|
|
|
this.parse_kv(res.responses[0].response_range.kvs[0]);
|
|
|
|
}
|
2020-05-14 19:29:35 +03:00
|
|
|
this.check_config();
|
|
|
|
}
|
|
|
|
|
|
|
|
check_config()
|
|
|
|
{
|
2020-09-01 02:13:03 +03:00
|
|
|
this.config.etcd_mon_ttl = Number(this.config.etcd_mon_ttl) || 30;
|
|
|
|
if (this.config.etcd_mon_ttl < 10)
|
|
|
|
{
|
|
|
|
this.config.etcd_mon_ttl = 10;
|
|
|
|
}
|
2020-05-12 17:49:18 +03:00
|
|
|
this.config.etcd_mon_timeout = Number(this.config.etcd_mon_timeout) || 0;
|
|
|
|
if (this.config.etcd_mon_timeout <= 0)
|
|
|
|
{
|
|
|
|
this.config.etcd_mon_timeout = 1000;
|
|
|
|
}
|
|
|
|
this.config.etcd_mon_retries = Number(this.config.etcd_mon_retries) || 5;
|
|
|
|
if (this.config.etcd_mon_retries < 0)
|
|
|
|
{
|
|
|
|
this.config.etcd_mon_retries = 0;
|
|
|
|
}
|
|
|
|
this.config.mon_change_timeout = Number(this.config.mon_change_timeout) || 1000;
|
|
|
|
if (this.config.mon_change_timeout < 100)
|
|
|
|
{
|
|
|
|
this.config.mon_change_timeout = 100;
|
|
|
|
}
|
2020-05-15 01:28:44 +03:00
|
|
|
this.config.mon_stats_timeout = Number(this.config.mon_stats_timeout) || 1000;
|
|
|
|
if (this.config.mon_stats_timeout < 100)
|
|
|
|
{
|
|
|
|
this.config.mon_stats_timeout = 100;
|
|
|
|
}
|
2020-05-12 17:49:18 +03:00
|
|
|
// After this number of seconds, a dead OSD will be removed from PG distribution
|
|
|
|
this.config.osd_out_time = Number(this.config.osd_out_time) || 0;
|
|
|
|
if (!this.config.osd_out_time)
|
|
|
|
{
|
2021-03-09 01:24:51 +03:00
|
|
|
this.config.osd_out_time = 600; // 10 minutes by default
|
2020-05-12 17:49:18 +03:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2021-11-28 01:19:42 +03:00
|
|
|
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.selected_etcd_url == cur_addr)
|
|
|
|
this.selected_etcd_url = null;
|
|
|
|
this.start_watcher(this.config.etcd_mon_retries).catch(this.die);
|
|
|
|
}
|
|
|
|
|
2020-05-12 17:49:18 +03:00
|
|
|
async start_watcher(retries)
|
|
|
|
{
|
|
|
|
let retry = 0;
|
2020-09-05 02:14:43 +03:00
|
|
|
if (!retries || retries < 1)
|
2020-05-12 17:49:18 +03:00
|
|
|
{
|
|
|
|
retries = 1;
|
|
|
|
}
|
|
|
|
while (retries < 0 || retry < retries)
|
|
|
|
{
|
2021-11-28 01:19:42 +03:00
|
|
|
const cur_addr = this.pick_next_etcd();
|
|
|
|
const base = 'ws'+cur_addr.substr(4);
|
2020-05-12 17:49:18 +03:00
|
|
|
const ok = await new Promise((ok, no) =>
|
|
|
|
{
|
|
|
|
const timer_id = setTimeout(() =>
|
|
|
|
{
|
|
|
|
this.ws.close();
|
|
|
|
ok(false);
|
2020-09-01 18:50:23 +03:00
|
|
|
}, this.config.etcd_mon_timeout);
|
2020-05-12 17:49:18 +03:00
|
|
|
this.ws = new WebSocket(base+'/watch');
|
2021-03-09 01:46:57 +03:00
|
|
|
const fail = () =>
|
|
|
|
{
|
|
|
|
ok(false);
|
|
|
|
};
|
|
|
|
this.ws.on('error', fail);
|
2020-05-12 17:49:18 +03:00
|
|
|
this.ws.on('open', () =>
|
|
|
|
{
|
2021-03-09 01:46:57 +03:00
|
|
|
this.ws.removeListener('error', fail);
|
2020-05-12 17:49:18 +03:00
|
|
|
if (timer_id)
|
|
|
|
clearTimeout(timer_id);
|
|
|
|
ok(true);
|
|
|
|
});
|
|
|
|
});
|
2020-09-05 02:14:43 +03:00
|
|
|
if (ok)
|
|
|
|
break;
|
2021-11-28 01:19:42 +03:00
|
|
|
if (this.selected_etcd_url == cur_addr)
|
|
|
|
this.selected_etcd_url = null;
|
2020-09-05 02:14:43 +03:00
|
|
|
this.ws = null;
|
2020-05-12 17:49:18 +03:00
|
|
|
retry++;
|
|
|
|
}
|
|
|
|
if (!this.ws)
|
|
|
|
{
|
|
|
|
this.die('Failed to open etcd watch websocket');
|
|
|
|
}
|
2021-11-28 01:19:42 +03:00
|
|
|
const cur_addr = this.selected_etcd_url;
|
|
|
|
this.ws.on('error', () => this.restart_watcher(cur_addr));
|
2020-05-12 17:49:18 +03:00
|
|
|
this.ws.send(JSON.stringify({
|
|
|
|
create_request: {
|
|
|
|
key: b64(this.etcd_prefix+'/'),
|
|
|
|
range_end: b64(this.etcd_prefix+'0'),
|
2020-05-14 19:29:35 +03:00
|
|
|
start_revision: ''+this.etcd_watch_revision,
|
2020-05-12 17:49:18 +03:00
|
|
|
watch_id: 1,
|
2020-11-17 16:29:42 +03:00
|
|
|
progress_notify: true,
|
2020-05-12 17:49:18 +03:00
|
|
|
},
|
|
|
|
}));
|
|
|
|
this.ws.on('message', (msg) =>
|
|
|
|
{
|
|
|
|
let data;
|
|
|
|
try
|
|
|
|
{
|
|
|
|
data = JSON.parse(msg);
|
|
|
|
}
|
|
|
|
catch (e)
|
|
|
|
{
|
|
|
|
}
|
2021-11-28 01:43:31 +03:00
|
|
|
if (!data || !data.result)
|
2020-05-12 17:49:18 +03:00
|
|
|
{
|
2021-11-28 01:43:31 +03:00
|
|
|
console.error('Unknown message received from watch websocket: '+msg);
|
|
|
|
}
|
|
|
|
else if (data.result.canceled)
|
|
|
|
{
|
|
|
|
// etcd watch canceled
|
|
|
|
if (data.result.compact_revision)
|
2020-09-05 02:14:43 +03:00
|
|
|
{
|
2021-11-28 01:43:31 +03:00
|
|
|
// we may miss events if we proceed
|
|
|
|
console.error('Revisions before '+data.result.compact_revision+' were compacted by etcd, exiting');
|
|
|
|
this.on_stop(1);
|
2020-09-05 02:14:43 +03:00
|
|
|
}
|
2021-11-28 01:43:31 +03:00
|
|
|
console.error('Watch canceled by etcd, reason: '+data.result.cancel_reason+', exiting');
|
|
|
|
this.on_stop(1);
|
|
|
|
}
|
|
|
|
else if (data.result.created)
|
|
|
|
{
|
|
|
|
// etcd watch created
|
2020-05-12 17:49:18 +03:00
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
2021-03-06 01:30:42 +03:00
|
|
|
let stats_changed = false, changed = false, pg_states_changed = false;
|
2020-09-11 18:49:03 +03:00
|
|
|
if (this.verbose)
|
|
|
|
{
|
|
|
|
console.log('Revision '+data.result.header.revision+' events: ');
|
|
|
|
}
|
2020-09-05 02:14:43 +03:00
|
|
|
this.etcd_watch_revision = BigInt(data.result.header.revision)+BigInt(1);
|
2020-05-12 17:49:18 +03:00
|
|
|
for (const e of data.result.events)
|
|
|
|
{
|
|
|
|
this.parse_kv(e.kv);
|
2020-05-14 19:29:35 +03:00
|
|
|
const key = e.kv.key.substr(this.etcd_prefix.length);
|
2021-01-21 00:30:18 +03:00
|
|
|
if (key.substr(0, 11) == '/osd/stats/' || key.substr(0, 10) == '/pg/stats/' || key.substr(0, 16) == '/osd/inodestats/')
|
2020-05-15 01:28:44 +03:00
|
|
|
{
|
|
|
|
stats_changed = true;
|
|
|
|
}
|
2021-03-06 01:30:42 +03:00
|
|
|
else if (key.substr(0, 10) == '/pg/state/')
|
|
|
|
{
|
|
|
|
pg_states_changed = true;
|
|
|
|
}
|
2021-01-21 00:30:18 +03:00
|
|
|
else if (key != '/stats' && key.substr(0, 13) != '/inode/stats/')
|
2020-05-14 19:29:35 +03:00
|
|
|
{
|
|
|
|
changed = true;
|
|
|
|
}
|
2020-09-11 18:49:03 +03:00
|
|
|
if (this.verbose)
|
|
|
|
{
|
2021-02-28 19:51:14 +03:00
|
|
|
console.log(JSON.stringify(e));
|
2020-09-11 18:49:03 +03:00
|
|
|
}
|
2020-05-12 17:49:18 +03:00
|
|
|
}
|
2021-03-06 01:30:42 +03:00
|
|
|
if (pg_states_changed)
|
|
|
|
{
|
2021-11-28 01:19:42 +03:00
|
|
|
this.save_last_clean().catch(this.die);
|
2021-03-06 01:30:42 +03:00
|
|
|
}
|
2020-05-15 01:28:44 +03:00
|
|
|
if (stats_changed)
|
|
|
|
{
|
|
|
|
this.schedule_update_stats();
|
|
|
|
}
|
2020-05-14 19:29:35 +03:00
|
|
|
if (changed)
|
2020-05-12 17:49:18 +03:00
|
|
|
{
|
2020-05-14 19:29:35 +03:00
|
|
|
this.schedule_recheck();
|
2020-05-12 17:49:18 +03:00
|
|
|
}
|
|
|
|
}
|
|
|
|
});
|
|
|
|
}
|
|
|
|
|
2021-03-06 01:30:42 +03:00
|
|
|
async save_last_clean()
|
|
|
|
{
|
|
|
|
// last_clean_pgs is used to avoid extra data move when observing a series of changes in the cluster
|
|
|
|
for (const pool_id in this.state.config.pools)
|
|
|
|
{
|
|
|
|
const pool_cfg = this.state.config.pools[pool_id];
|
|
|
|
if (!this.validate_pool_cfg(pool_id, pool_cfg, false))
|
|
|
|
{
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
for (let pg_num = 1; pg_num <= pool_cfg.pg_count; pg_num++)
|
|
|
|
{
|
|
|
|
if (!this.state.pg.state[pool_id] ||
|
|
|
|
!this.state.pg.state[pool_id][pg_num] ||
|
|
|
|
!(this.state.pg.state[pool_id][pg_num].state instanceof Array))
|
|
|
|
{
|
|
|
|
// Unclean
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
let st = this.state.pg.state[pool_id][pg_num].state.join(',');
|
|
|
|
if (st != 'active' && st != 'active,left_on_dead' && st != 'left_on_dead,active')
|
|
|
|
{
|
|
|
|
// Unclean
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
this.state.history.last_clean_pgs = JSON.parse(JSON.stringify(this.state.config.pgs));
|
|
|
|
await this.etcd_call('/kv/txn', {
|
|
|
|
success: [ { requestPut: {
|
|
|
|
key: b64(this.etcd_prefix+'/history/last_clean_pgs'),
|
|
|
|
value: b64(JSON.stringify(this.state.history.last_clean_pgs))
|
|
|
|
} } ],
|
|
|
|
}, this.etcd_start_timeout, 0);
|
|
|
|
}
|
|
|
|
|
2020-05-12 17:49:18 +03:00
|
|
|
async get_lease()
|
|
|
|
{
|
|
|
|
const max_ttl = this.config.etcd_mon_ttl + this.config.etcd_mon_timeout/1000*this.config.etcd_mon_retries;
|
2021-03-09 01:46:57 +03:00
|
|
|
const res = await this.etcd_call('/lease/grant', { TTL: max_ttl }, this.config.etcd_mon_timeout, -1);
|
2020-05-12 17:49:18 +03:00
|
|
|
this.etcd_lease_id = res.ID;
|
2021-11-06 13:38:52 +03:00
|
|
|
this.lease_timer = setInterval(async () =>
|
2020-05-12 17:49:18 +03:00
|
|
|
{
|
|
|
|
const res = await this.etcd_call('/lease/keepalive', { ID: this.etcd_lease_id }, this.config.etcd_mon_timeout, this.config.etcd_mon_retries);
|
|
|
|
if (!res.result.TTL)
|
|
|
|
{
|
|
|
|
this.die('Lease expired');
|
|
|
|
}
|
2020-09-01 18:50:23 +03:00
|
|
|
}, this.config.etcd_mon_timeout);
|
2021-11-06 13:38:52 +03:00
|
|
|
if (!this.signals_set)
|
|
|
|
{
|
|
|
|
process.on('SIGINT', this.on_stop_cb);
|
|
|
|
process.on('SIGTERM', this.on_stop_cb);
|
|
|
|
this.signals_set = true;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2021-11-28 01:43:31 +03:00
|
|
|
async on_stop(status)
|
2021-11-06 13:38:52 +03:00
|
|
|
{
|
|
|
|
clearInterval(this.lease_timer);
|
|
|
|
await this.etcd_call('/lease/revoke', { ID: this.etcd_lease_id }, this.config.etcd_mon_timeout, this.config.etcd_mon_retries);
|
2021-11-28 01:43:31 +03:00
|
|
|
process.exit(status);
|
2020-05-12 17:49:18 +03:00
|
|
|
}
|
|
|
|
|
|
|
|
async become_master()
|
|
|
|
{
|
|
|
|
const state = { ip: this.local_ips() };
|
|
|
|
while (1)
|
|
|
|
{
|
2020-09-05 02:14:43 +03:00
|
|
|
const res = await this.etcd_call('/kv/txn', {
|
2020-05-12 17:49:18 +03:00
|
|
|
compare: [ { target: 'CREATE', create_revision: 0, key: b64(this.etcd_prefix+'/mon/master') } ],
|
2020-09-05 02:14:43 +03:00
|
|
|
success: [ { requestPut: { key: b64(this.etcd_prefix+'/mon/master'), value: b64(JSON.stringify(state)), lease: ''+this.etcd_lease_id } } ],
|
2020-05-12 17:49:18 +03:00
|
|
|
}, this.etcd_start_timeout, 0);
|
2020-09-05 02:14:43 +03:00
|
|
|
if (res.succeeded)
|
2020-05-12 17:49:18 +03:00
|
|
|
{
|
2020-09-05 02:14:43 +03:00
|
|
|
break;
|
2020-05-12 17:49:18 +03:00
|
|
|
}
|
2020-09-05 02:14:43 +03:00
|
|
|
console.log('Waiting to become master');
|
|
|
|
await new Promise(ok => setTimeout(ok, this.etcd_start_timeout));
|
2020-05-12 17:49:18 +03:00
|
|
|
}
|
2020-09-05 02:14:43 +03:00
|
|
|
console.log('Became master');
|
2020-05-12 17:49:18 +03:00
|
|
|
}
|
|
|
|
|
|
|
|
async load_cluster_state()
|
|
|
|
{
|
2020-09-05 02:14:43 +03:00
|
|
|
const res = await this.etcd_call('/kv/txn', { success: [
|
2020-05-12 17:49:18 +03:00
|
|
|
{ requestRange: { key: b64(this.etcd_prefix+'/'), range_end: b64(this.etcd_prefix+'0') } },
|
|
|
|
] }, this.etcd_start_timeout, -1);
|
|
|
|
this.etcd_watch_revision = BigInt(res.header.revision)+BigInt(1);
|
2020-09-05 02:14:43 +03:00
|
|
|
this.state = JSON.parse(JSON.stringify(this.constructor.etcd_tree));
|
2020-05-12 17:49:18 +03:00
|
|
|
for (const response of res.responses)
|
|
|
|
{
|
|
|
|
for (const kv of response.response_range.kvs)
|
|
|
|
{
|
|
|
|
this.parse_kv(kv);
|
|
|
|
}
|
|
|
|
}
|
2020-05-14 19:29:35 +03:00
|
|
|
}
|
|
|
|
|
|
|
|
all_osds()
|
|
|
|
{
|
|
|
|
return Object.keys(this.state.osd.stats);
|
2020-05-12 17:49:18 +03:00
|
|
|
}
|
|
|
|
|
|
|
|
get_osd_tree()
|
|
|
|
{
|
2020-09-01 02:13:03 +03:00
|
|
|
const levels = this.config.placement_levels||{};
|
2020-05-12 17:49:18 +03:00
|
|
|
levels.host = levels.host || 100;
|
|
|
|
levels.osd = levels.osd || 101;
|
|
|
|
const tree = { '': { children: [] } };
|
2020-09-13 12:23:11 +03:00
|
|
|
let up_osds = {};
|
2020-09-01 02:13:03 +03:00
|
|
|
for (const node_id in this.state.config.node_placement||{})
|
2020-05-12 17:49:18 +03:00
|
|
|
{
|
2020-09-01 02:13:03 +03:00
|
|
|
const node_cfg = this.state.config.node_placement[node_id];
|
2021-12-12 01:25:45 +03:00
|
|
|
if (/^\d+$/.exec(node_id))
|
2020-05-12 17:49:18 +03:00
|
|
|
{
|
2021-12-12 01:25:45 +03:00
|
|
|
node_cfg.level = 'osd';
|
|
|
|
}
|
|
|
|
if (!node_id || !node_cfg.level || !levels[node_cfg.level])
|
|
|
|
{
|
|
|
|
// All nodes must have non-empty IDs and valid levels
|
2020-05-12 17:49:18 +03:00
|
|
|
continue;
|
|
|
|
}
|
|
|
|
tree[node_id] = { id: node_id, level: node_cfg.level, parent: node_cfg.parent, children: [] };
|
|
|
|
}
|
2020-05-14 19:29:35 +03:00
|
|
|
// This requires monitor system time to be in sync with OSD system times (at least to some extent)
|
2020-05-12 17:49:18 +03:00
|
|
|
const down_time = Date.now()/1000 - this.config.osd_out_time;
|
2020-05-14 19:29:35 +03:00
|
|
|
for (const osd_num of this.all_osds().sort((a, b) => a - b))
|
2020-05-12 17:49:18 +03:00
|
|
|
{
|
2020-05-14 19:29:35 +03:00
|
|
|
const stat = this.state.osd.stats[osd_num];
|
2021-03-27 01:15:05 +03:00
|
|
|
if (stat && stat.size && (this.state.osd.state[osd_num] || Number(stat.time) >= down_time))
|
2020-05-12 17:49:18 +03:00
|
|
|
{
|
|
|
|
// Numeric IDs are reserved for OSDs
|
2020-10-04 17:20:09 +03:00
|
|
|
const osd_cfg = this.state.config.osd[osd_num];
|
|
|
|
let reweight = osd_cfg && Number(osd_cfg.reweight);
|
2020-09-11 23:59:30 +03:00
|
|
|
if (reweight < 0 || isNaN(reweight))
|
|
|
|
reweight = 1;
|
2020-09-13 12:23:11 +03:00
|
|
|
if (this.state.osd.state[osd_num] && reweight > 0)
|
|
|
|
{
|
|
|
|
// React to down OSDs immediately
|
|
|
|
up_osds[osd_num] = true;
|
|
|
|
}
|
2020-10-21 00:07:40 +03:00
|
|
|
tree[osd_num] = tree[osd_num] || {};
|
|
|
|
tree[osd_num].id = osd_num;
|
|
|
|
tree[osd_num].parent = tree[osd_num].parent || stat.host;
|
2020-05-12 17:49:18 +03:00
|
|
|
tree[osd_num].level = 'osd';
|
2020-05-22 12:52:27 +03:00
|
|
|
tree[osd_num].size = reweight * stat.size / 1024 / 1024 / 1024 / 1024; // terabytes
|
2020-10-04 17:20:09 +03:00
|
|
|
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; }, {});
|
|
|
|
}
|
2020-05-12 17:49:18 +03:00
|
|
|
delete tree[osd_num].children;
|
2021-12-12 01:25:45 +03:00
|
|
|
if (!tree[stat.host])
|
2020-10-21 00:07:40 +03:00
|
|
|
{
|
2021-12-12 01:25:45 +03:00
|
|
|
tree[stat.host] = {
|
|
|
|
id: stat.host,
|
2020-10-21 00:07:40 +03:00
|
|
|
level: 'host',
|
|
|
|
parent: null,
|
|
|
|
children: [],
|
|
|
|
};
|
|
|
|
}
|
2020-05-12 17:49:18 +03:00
|
|
|
}
|
|
|
|
}
|
|
|
|
for (const node_id in tree)
|
|
|
|
{
|
|
|
|
if (node_id === '')
|
|
|
|
{
|
|
|
|
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
|
2020-10-21 00:07:40 +03:00
|
|
|
const parent = parent_level && parent_level < node_level ? node_cfg.parent : '';
|
2020-05-12 17:49:18 +03:00
|
|
|
tree[parent].children.push(tree[node_id]);
|
|
|
|
delete node_cfg.parent;
|
|
|
|
}
|
2020-10-04 17:20:09 +03:00
|
|
|
return { up_osds, levels, osd_tree: tree };
|
2020-05-12 17:49:18 +03:00
|
|
|
}
|
|
|
|
|
2020-09-01 18:50:23 +03:00
|
|
|
async stop_all_pgs(pool_id)
|
2020-05-12 17:49:18 +03:00
|
|
|
{
|
2020-05-14 19:29:35 +03:00
|
|
|
let has_online = false, paused = true;
|
2020-09-01 18:50:23 +03:00
|
|
|
for (const pg in this.state.config.pgs.items[pool_id]||{})
|
2020-05-12 17:49:18 +03:00
|
|
|
{
|
2020-09-01 18:50:23 +03:00
|
|
|
// FIXME: Change all (||{}) to ?. (optional chaining) at some point
|
|
|
|
const cur_state = (((this.state.pg.state[pool_id]||{})[pg]||{}).state||[]).join(',');
|
2020-05-14 19:29:35 +03:00
|
|
|
if (cur_state != '' && cur_state != 'offline')
|
|
|
|
{
|
|
|
|
has_online = true;
|
|
|
|
}
|
2020-09-01 18:50:23 +03:00
|
|
|
if (!this.state.config.pgs.items[pool_id][pg].pause)
|
2020-05-14 19:29:35 +03:00
|
|
|
{
|
|
|
|
paused = false;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
if (!paused)
|
|
|
|
{
|
2020-09-01 18:50:23 +03:00
|
|
|
console.log('Stopping all PGs for pool '+pool_id+' before changing PG count');
|
2020-05-14 19:29:35 +03:00
|
|
|
const new_cfg = JSON.parse(JSON.stringify(this.state.config.pgs));
|
2020-09-01 18:50:23 +03:00
|
|
|
for (const pg in new_cfg.items[pool_id])
|
2020-05-14 19:29:35 +03:00
|
|
|
{
|
2020-09-01 18:50:23 +03:00
|
|
|
new_cfg.items[pool_id][pg].pause = true;
|
2020-05-14 19:29:35 +03:00
|
|
|
}
|
|
|
|
// Check that no OSDs change their state before we pause PGs
|
|
|
|
// Doing this we make sure that OSDs don't wake up in the middle of our "transaction"
|
|
|
|
// and can't see the old PG configuration
|
|
|
|
const checks = [];
|
|
|
|
for (const osd_num of this.all_osds())
|
|
|
|
{
|
|
|
|
const key = b64(this.etcd_prefix+'/osd/state/'+osd_num);
|
|
|
|
checks.push({ key, target: 'MOD', result: 'LESS', mod_revision: ''+this.etcd_watch_revision });
|
|
|
|
}
|
2020-09-05 02:14:43 +03:00
|
|
|
const res = await this.etcd_call('/kv/txn', {
|
2020-05-14 19:29:35 +03:00
|
|
|
compare: [
|
|
|
|
{ key: b64(this.etcd_prefix+'/mon/master'), target: 'LEASE', lease: ''+this.etcd_lease_id },
|
|
|
|
{ key: b64(this.etcd_prefix+'/config/pgs'), target: 'MOD', mod_revision: ''+this.etcd_watch_revision, result: 'LESS' },
|
|
|
|
...checks,
|
|
|
|
],
|
|
|
|
success: [
|
|
|
|
{ requestPut: { key: b64(this.etcd_prefix+'/config/pgs'), value: b64(JSON.stringify(new_cfg)) } },
|
|
|
|
],
|
|
|
|
}, this.config.etcd_mon_timeout, 0);
|
2020-12-01 14:13:24 +03:00
|
|
|
return false;
|
2020-05-14 19:29:35 +03:00
|
|
|
}
|
|
|
|
return !has_online;
|
|
|
|
}
|
|
|
|
|
2021-03-14 01:26:06 +03:00
|
|
|
reset_rng()
|
|
|
|
{
|
|
|
|
this.seed = 0x5f020e43;
|
|
|
|
}
|
|
|
|
|
|
|
|
rng()
|
|
|
|
{
|
|
|
|
this.seed ^= this.seed << 13;
|
|
|
|
this.seed ^= this.seed >> 17;
|
|
|
|
this.seed ^= this.seed << 5;
|
|
|
|
return this.seed + 2147483648;
|
|
|
|
}
|
|
|
|
|
|
|
|
pick_primary(pool_id, osd_set, up_osds)
|
|
|
|
{
|
|
|
|
let alive_set;
|
|
|
|
if (this.state.config.pools[pool_id].scheme === 'replicated')
|
|
|
|
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 = (this.state.config.pools[pool_id].pg_size||0) -
|
|
|
|
(this.state.config.pools[pool_id].parity_chunks||0);
|
|
|
|
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[this.rng() % alive_set.length];
|
|
|
|
}
|
|
|
|
|
2020-09-16 01:34:18 +03:00
|
|
|
save_new_pgs_txn(request, pool_id, up_osds, prev_pgs, new_pgs, pg_history)
|
2020-05-14 19:29:35 +03:00
|
|
|
{
|
|
|
|
const pg_items = {};
|
2021-03-14 01:26:06 +03:00
|
|
|
this.reset_rng();
|
2020-05-14 19:29:35 +03:00
|
|
|
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,
|
2021-03-14 01:26:06 +03:00
|
|
|
primary: this.pick_primary(pool_id, osd_set, up_osds),
|
2020-05-14 19:29:35 +03:00
|
|
|
};
|
2020-03-12 00:02:18 +03:00
|
|
|
if (prev_pgs[i] && prev_pgs[i].join(' ') != osd_set.join(' ') &&
|
|
|
|
prev_pgs[i].filter(osd_num => osd_num).length > 0)
|
2020-05-14 19:29:35 +03:00
|
|
|
{
|
|
|
|
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]);
|
|
|
|
}
|
2021-03-27 18:16:00 +03:00
|
|
|
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; }, {}));
|
|
|
|
}
|
2020-05-14 19:29:35 +03:00
|
|
|
});
|
|
|
|
for (let i = 0; i < new_pgs.length || i < prev_pgs.length; i++)
|
|
|
|
{
|
2020-09-13 12:23:11 +03:00
|
|
|
// 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...
|
2020-09-01 18:50:23 +03:00
|
|
|
request.compare.push({
|
|
|
|
key: b64(this.etcd_prefix+'/pg/history/'+pool_id+'/'+(i+1)),
|
2020-05-14 19:29:35 +03:00
|
|
|
target: 'MOD',
|
|
|
|
mod_revision: ''+this.etcd_watch_revision,
|
|
|
|
result: 'LESS',
|
2020-05-12 17:49:18 +03:00
|
|
|
});
|
2020-05-14 19:29:35 +03:00
|
|
|
if (pg_history[i])
|
2020-05-12 17:49:18 +03:00
|
|
|
{
|
2020-09-01 18:50:23 +03:00
|
|
|
request.success.push({
|
2020-05-14 19:29:35 +03:00
|
|
|
requestPut: {
|
2020-09-01 18:50:23 +03:00
|
|
|
key: b64(this.etcd_prefix+'/pg/history/'+pool_id+'/'+(i+1)),
|
2020-05-14 19:29:35 +03:00
|
|
|
value: b64(JSON.stringify(pg_history[i])),
|
|
|
|
},
|
|
|
|
});
|
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
2020-09-01 18:50:23 +03:00
|
|
|
request.success.push({
|
2020-05-14 19:29:35 +03:00
|
|
|
requestDeleteRange: {
|
2020-09-01 18:50:23 +03:00
|
|
|
key: b64(this.etcd_prefix+'/pg/history/'+pool_id+'/'+(i+1)),
|
2020-05-14 19:29:35 +03:00
|
|
|
},
|
|
|
|
});
|
|
|
|
}
|
|
|
|
}
|
2020-09-05 02:14:43 +03:00
|
|
|
this.state.config.pgs.items = this.state.config.pgs.items || {};
|
2020-12-01 14:13:24 +03:00
|
|
|
if (!new_pgs.length)
|
|
|
|
{
|
|
|
|
delete this.state.config.pgs.items[pool_id];
|
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
|
|
|
this.state.config.pgs.items[pool_id] = pg_items;
|
|
|
|
}
|
2020-09-01 18:50:23 +03:00
|
|
|
}
|
|
|
|
|
2020-09-05 02:14:43 +03:00
|
|
|
validate_pool_cfg(pool_id, pool_cfg, warn)
|
2020-09-01 18:50:23 +03:00
|
|
|
{
|
|
|
|
pool_cfg.pg_size = Math.floor(pool_cfg.pg_size);
|
|
|
|
pool_cfg.pg_minsize = Math.floor(pool_cfg.pg_minsize);
|
2020-11-30 00:08:25 +03:00
|
|
|
pool_cfg.parity_chunks = Math.floor(pool_cfg.parity_chunks) || undefined;
|
2020-09-01 18:50:23 +03:00
|
|
|
pool_cfg.pg_count = Math.floor(pool_cfg.pg_count);
|
|
|
|
pool_cfg.failure_domain = pool_cfg.failure_domain || 'host';
|
|
|
|
pool_cfg.max_osd_combinations = Math.floor(pool_cfg.max_osd_combinations) || 10000;
|
|
|
|
if (!/^[1-9]\d*$/.exec(''+pool_id))
|
|
|
|
{
|
2020-09-05 02:14:43 +03:00
|
|
|
if (warn)
|
|
|
|
console.log('Pool ID '+pool_id+' is invalid');
|
2020-09-01 18:50:23 +03:00
|
|
|
return false;
|
|
|
|
}
|
2020-11-30 00:08:25 +03:00
|
|
|
if (pool_cfg.scheme !== 'xor' && pool_cfg.scheme !== 'replicated' && pool_cfg.scheme !== 'jerasure')
|
|
|
|
{
|
|
|
|
if (warn)
|
|
|
|
console.log('Pool '+pool_id+' has invalid coding scheme (one of "xor", "replicated" and "jerasure" required)');
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
if (!pool_cfg.pg_size || pool_cfg.pg_size < 1 || pool_cfg.pg_size > 256 ||
|
|
|
|
(pool_cfg.scheme === 'xor' || pool_cfg.scheme == 'jerasure') && pool_cfg.pg_size < 3)
|
2020-09-01 18:50:23 +03:00
|
|
|
{
|
2020-09-05 02:14:43 +03:00
|
|
|
if (warn)
|
|
|
|
console.log('Pool '+pool_id+' has invalid pg_size');
|
2020-09-01 18:50:23 +03:00
|
|
|
return false;
|
|
|
|
}
|
2020-09-02 21:54:32 +03:00
|
|
|
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))
|
2020-09-01 18:50:23 +03:00
|
|
|
{
|
2020-09-05 02:14:43 +03:00
|
|
|
if (warn)
|
|
|
|
console.log('Pool '+pool_id+' has invalid pg_minsize');
|
2020-09-01 18:50:23 +03:00
|
|
|
return false;
|
|
|
|
}
|
2020-11-30 00:08:25 +03:00
|
|
|
if (pool_cfg.scheme === 'xor' && pool_cfg.parity_chunks != 0 && pool_cfg.parity_chunks != 1)
|
2020-09-01 18:50:23 +03:00
|
|
|
{
|
2020-09-05 02:14:43 +03:00
|
|
|
if (warn)
|
2020-11-30 00:08:25 +03:00
|
|
|
console.log('Pool '+pool_id+' has invalid parity_chunks (must be 1)');
|
2020-09-01 18:50:23 +03:00
|
|
|
return false;
|
|
|
|
}
|
2020-11-30 00:08:25 +03:00
|
|
|
if (pool_cfg.scheme === 'jerasure' && (pool_cfg.parity_chunks < 1 || pool_cfg.parity_chunks > pool_cfg.pg_size-2))
|
2020-09-01 18:50:23 +03:00
|
|
|
{
|
2020-09-05 02:14:43 +03:00
|
|
|
if (warn)
|
2020-11-30 00:08:25 +03:00
|
|
|
console.log('Pool '+pool_id+' has invalid parity_chunks (must be between 1 and pg_size-2)');
|
2020-09-02 21:54:32 +03:00
|
|
|
return false;
|
|
|
|
}
|
2020-11-30 00:08:25 +03:00
|
|
|
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)
|
2020-09-02 21:54:32 +03:00
|
|
|
{
|
2020-09-05 02:14:43 +03:00
|
|
|
if (warn)
|
2020-11-30 00:08:25 +03:00
|
|
|
console.log('Pool '+pool_id+' has empty name');
|
2020-09-01 18:50:23 +03:00
|
|
|
return false;
|
|
|
|
}
|
|
|
|
if (pool_cfg.max_osd_combinations < 100)
|
|
|
|
{
|
2020-09-05 02:14:43 +03:00
|
|
|
if (warn)
|
|
|
|
console.log('Pool '+pool_id+' has invalid max_osd_combinations (must be at least 100)');
|
2020-09-01 18:50:23 +03:00
|
|
|
return false;
|
|
|
|
}
|
2020-10-04 17:20:09 +03:00
|
|
|
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;
|
|
|
|
}
|
2020-09-01 18:50:23 +03:00
|
|
|
return true;
|
2020-05-14 19:29:35 +03:00
|
|
|
}
|
|
|
|
|
2020-10-04 17:20:09 +03:00
|
|
|
filter_osds_by_tags(orig_tree, flat_tree, tags)
|
|
|
|
{
|
|
|
|
if (!tags)
|
|
|
|
{
|
2020-10-17 02:28:48 +03:00
|
|
|
return;
|
2020-10-04 17:20:09 +03:00
|
|
|
}
|
|
|
|
for (const tag of (tags instanceof Array ? tags : [ tags ]))
|
|
|
|
{
|
|
|
|
for (const host in flat_tree)
|
|
|
|
{
|
|
|
|
let found = 0;
|
|
|
|
for (const osd in flat_tree[host])
|
|
|
|
{
|
|
|
|
if (!orig_tree[osd].tags || !orig_tree[osd].tags[tag])
|
|
|
|
delete flat_tree[host][osd];
|
|
|
|
else
|
|
|
|
found++;
|
|
|
|
}
|
|
|
|
if (!found)
|
|
|
|
{
|
|
|
|
delete flat_tree[host];
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2020-05-14 19:29:35 +03:00
|
|
|
async recheck_pgs()
|
|
|
|
{
|
|
|
|
// Take configuration and state, check it against the stored configuration hash
|
|
|
|
// Recalculate PGs and save them to etcd if the configuration is changed
|
2021-04-03 23:39:25 +03:00
|
|
|
// FIXME: Do not change anything if the distribution is good and random enough and no PGs are degraded
|
2020-10-04 17:20:09 +03:00
|
|
|
const { up_osds, levels, osd_tree } = this.get_osd_tree();
|
2020-05-14 19:29:35 +03:00
|
|
|
const tree_cfg = {
|
2020-09-13 12:23:11 +03:00
|
|
|
osd_tree,
|
2020-09-01 18:50:23 +03:00
|
|
|
pools: this.state.config.pools,
|
2020-05-14 19:29:35 +03:00
|
|
|
};
|
|
|
|
const tree_hash = sha1hex(stableStringify(tree_cfg));
|
|
|
|
if (this.state.config.pgs.hash != tree_hash)
|
|
|
|
{
|
|
|
|
// Something has changed
|
2020-09-01 18:50:23 +03:00
|
|
|
const etcd_request = { compare: [], success: [] };
|
2020-12-01 14:13:24 +03:00
|
|
|
for (const pool_id in (this.state.config.pgs||{}).items||{})
|
|
|
|
{
|
|
|
|
if (!this.state.config.pools[pool_id])
|
|
|
|
{
|
|
|
|
// Pool deleted. Delete all PGs, but first stop them.
|
|
|
|
if (!await this.stop_all_pgs(pool_id))
|
|
|
|
{
|
|
|
|
this.schedule_recheck();
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
const prev_pgs = [];
|
|
|
|
for (const pg in this.state.config.pgs.items[pool_id]||{})
|
|
|
|
{
|
|
|
|
prev_pgs[pg-1] = this.state.config.pgs.items[pool_id][pg].osd_set;
|
|
|
|
}
|
|
|
|
this.save_new_pgs_txn(etcd_request, pool_id, up_osds, prev_pgs, [], []);
|
|
|
|
}
|
|
|
|
}
|
2020-09-01 18:50:23 +03:00
|
|
|
for (const pool_id in this.state.config.pools)
|
2020-05-14 19:29:35 +03:00
|
|
|
{
|
2020-09-01 18:50:23 +03:00
|
|
|
const pool_cfg = this.state.config.pools[pool_id];
|
2020-09-05 02:14:43 +03:00
|
|
|
if (!this.validate_pool_cfg(pool_id, pool_cfg, false))
|
2020-09-01 18:50:23 +03:00
|
|
|
{
|
2020-09-05 02:14:43 +03:00
|
|
|
continue;
|
2020-09-01 18:50:23 +03:00
|
|
|
}
|
2020-10-04 17:20:09 +03:00
|
|
|
let pool_tree = osd_tree[pool_cfg.root_node || ''];
|
|
|
|
pool_tree = pool_tree ? pool_tree.children : [];
|
|
|
|
pool_tree = LPOptimizer.flatten_tree(pool_tree, levels, pool_cfg.failure_domain, 'osd');
|
2020-10-17 02:28:48 +03:00
|
|
|
this.filter_osds_by_tags(osd_tree, pool_tree, pool_cfg.osd_tags);
|
2021-03-10 01:13:05 +03:00
|
|
|
// These are for the purpose of building history.osd_sets
|
|
|
|
const real_prev_pgs = [];
|
2021-03-11 00:12:55 +03:00
|
|
|
let pg_history = [];
|
2021-03-10 01:13:05 +03:00
|
|
|
for (const pg in ((this.state.config.pgs.items||{})[pool_id]||{}))
|
2020-09-01 18:50:23 +03:00
|
|
|
{
|
2021-03-10 01:13:05 +03:00
|
|
|
real_prev_pgs[pg-1] = this.state.config.pgs.items[pool_id][pg].osd_set;
|
2021-03-11 00:12:55 +03:00
|
|
|
if (this.state.pg.history[pool_id] &&
|
|
|
|
this.state.pg.history[pool_id][pg])
|
|
|
|
{
|
|
|
|
pg_history[pg-1] = this.state.pg.history[pool_id][pg];
|
|
|
|
}
|
2020-09-01 18:50:23 +03:00
|
|
|
}
|
2021-03-10 01:13:05 +03:00
|
|
|
// And these are for the purpose of minimizing data movement
|
|
|
|
let prev_pgs = [];
|
|
|
|
for (const pg in ((this.state.history.last_clean_pgs.items||{})[pool_id]||{}))
|
|
|
|
{
|
|
|
|
prev_pgs[pg-1] = this.state.history.last_clean_pgs.items[pool_id][pg].osd_set;
|
|
|
|
}
|
|
|
|
prev_pgs = JSON.parse(JSON.stringify(prev_pgs.length ? prev_pgs : real_prev_pgs));
|
2021-04-03 23:39:25 +03:00
|
|
|
const old_pg_count = real_prev_pgs.length;
|
|
|
|
const optimize_cfg = {
|
|
|
|
osd_tree: pool_tree,
|
|
|
|
pg_count: pool_cfg.pg_count,
|
|
|
|
pg_size: pool_cfg.pg_size,
|
|
|
|
pg_minsize: pool_cfg.pg_minsize,
|
|
|
|
max_combinations: pool_cfg.max_osd_combinations,
|
2021-11-16 21:02:33 +03:00
|
|
|
round_robin: pool_cfg.scheme != 'replicated',
|
2021-04-03 23:39:25 +03:00
|
|
|
};
|
2020-09-01 18:50:23 +03:00
|
|
|
let optimize_result;
|
|
|
|
if (old_pg_count > 0)
|
2020-05-12 17:49:18 +03:00
|
|
|
{
|
2020-09-01 18:50:23 +03:00
|
|
|
if (old_pg_count != pool_cfg.pg_count)
|
2020-05-14 19:29:35 +03:00
|
|
|
{
|
2020-09-01 18:50:23 +03:00
|
|
|
// PG count changed. Need to bring all PGs down.
|
|
|
|
if (!await this.stop_all_pgs(pool_id))
|
|
|
|
{
|
|
|
|
this.schedule_recheck();
|
|
|
|
return;
|
|
|
|
}
|
2021-03-11 00:12:55 +03:00
|
|
|
const new_pg_history = [];
|
|
|
|
PGUtil.scale_pg_count(prev_pgs, pg_history, new_pg_history, pool_cfg.pg_count);
|
|
|
|
pg_history = new_pg_history;
|
2020-05-14 19:29:35 +03:00
|
|
|
}
|
2021-02-25 23:21:56 +03:00
|
|
|
for (const pg of prev_pgs)
|
|
|
|
{
|
|
|
|
while (pg.length < pool_cfg.pg_size)
|
|
|
|
{
|
|
|
|
pg.push(0);
|
|
|
|
}
|
|
|
|
while (pg.length > pool_cfg.pg_size)
|
|
|
|
{
|
|
|
|
pg.pop();
|
|
|
|
}
|
|
|
|
}
|
2021-04-03 23:39:25 +03:00
|
|
|
if (!this.state.config.pgs.hash)
|
|
|
|
{
|
|
|
|
// Re-shuffle PGs
|
|
|
|
optimize_result = await LPOptimizer.optimize_initial(optimize_cfg);
|
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
|
|
|
optimize_result = await LPOptimizer.optimize_change({
|
|
|
|
prev_pgs,
|
|
|
|
...optimize_cfg,
|
|
|
|
});
|
|
|
|
}
|
2020-05-12 17:49:18 +03:00
|
|
|
}
|
2020-09-01 18:50:23 +03:00
|
|
|
else
|
|
|
|
{
|
2021-04-03 23:39:25 +03:00
|
|
|
optimize_result = await LPOptimizer.optimize_initial(optimize_cfg);
|
2020-09-01 18:50:23 +03:00
|
|
|
}
|
|
|
|
if (old_pg_count != optimize_result.int_pgs.length)
|
|
|
|
{
|
|
|
|
console.log(
|
2020-09-13 12:23:11 +03:00
|
|
|
`PG count for pool ${pool_id} (${pool_cfg.name || 'unnamed'})`+
|
|
|
|
` changed from: ${old_pg_count} to ${optimize_result.int_pgs.length}`
|
2020-09-01 18:50:23 +03:00
|
|
|
);
|
2021-03-08 23:13:08 +03:00
|
|
|
// Drop stats
|
|
|
|
etcd_request.success.push({ requestDeleteRange: {
|
|
|
|
key: b64(this.etcd_prefix+'/pg/stats/'+pool_id+'/'),
|
|
|
|
range_end: b64(this.etcd_prefix+'/pg/stats/'+pool_id+'0'),
|
|
|
|
} });
|
2020-09-01 18:50:23 +03:00
|
|
|
}
|
|
|
|
LPOptimizer.print_change_stats(optimize_result);
|
2021-07-02 22:47:01 +03:00
|
|
|
const pg_effsize = Math.min(pool_cfg.pg_size, Object.keys(pool_tree).length);
|
|
|
|
this.state.pool.stats[pool_id] = {
|
|
|
|
used_raw_tb: (this.state.pool.stats[pool_id]||{}).used_raw_tb || 0,
|
|
|
|
total_raw_tb: optimize_result.space,
|
2021-11-06 13:38:52 +03:00
|
|
|
pg_real_size: pg_effsize,
|
|
|
|
raw_to_usable: pg_effsize / (pool_cfg.scheme === 'replicated'
|
|
|
|
? 1 : (pool_cfg.pg_size - (pool_cfg.parity_chunks||0))),
|
2021-07-02 22:47:01 +03:00
|
|
|
space_efficiency: optimize_result.space/(optimize_result.total_space||1),
|
|
|
|
};
|
|
|
|
etcd_request.success.push({ requestPut: {
|
|
|
|
key: b64(this.etcd_prefix+'/pool/stats/'+pool_id),
|
|
|
|
value: b64(JSON.stringify(this.state.pool.stats[pool_id])),
|
|
|
|
} });
|
2021-03-10 01:13:05 +03:00
|
|
|
this.save_new_pgs_txn(etcd_request, pool_id, up_osds, real_prev_pgs, optimize_result.int_pgs, pg_history);
|
2020-05-14 19:29:35 +03:00
|
|
|
}
|
2020-09-01 18:50:23 +03:00
|
|
|
this.state.config.pgs.hash = tree_hash;
|
2021-02-25 20:36:37 +03:00
|
|
|
await this.save_pg_config(etcd_request);
|
2020-09-13 12:23:11 +03:00
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
2021-03-14 01:26:06 +03:00
|
|
|
// Nothing changed, but we still want to recheck the distribution of primaries
|
2020-09-13 12:23:11 +03:00
|
|
|
let changed = false;
|
|
|
|
for (const pool_id in this.state.config.pools)
|
2020-05-14 19:29:35 +03:00
|
|
|
{
|
2020-09-13 12:23:11 +03:00
|
|
|
const pool_cfg = this.state.config.pools[pool_id];
|
|
|
|
if (!this.validate_pool_cfg(pool_id, pool_cfg, false))
|
|
|
|
{
|
|
|
|
continue;
|
|
|
|
}
|
2020-09-16 01:34:18 +03:00
|
|
|
const replicated = pool_cfg.scheme === 'replicated';
|
2021-03-14 01:26:06 +03:00
|
|
|
this.reset_rng();
|
|
|
|
for (let pg_num = 1; pg_num <= pool_cfg.pg_count; pg_num++)
|
2020-09-13 12:23:11 +03:00
|
|
|
{
|
2020-10-17 02:28:48 +03:00
|
|
|
const pg_cfg = this.state.config.pgs.items[pool_id][pg_num];
|
2021-03-14 01:26:06 +03:00
|
|
|
if (pg_cfg)
|
2020-09-13 12:23:11 +03:00
|
|
|
{
|
2021-03-14 01:26:06 +03:00
|
|
|
const new_primary = this.pick_primary(pool_id, pg_cfg.osd_set, up_osds);
|
2020-09-13 12:23:11 +03:00
|
|
|
if (pg_cfg.primary != new_primary)
|
|
|
|
{
|
|
|
|
console.log(
|
|
|
|
`Moving pool ${pool_id} (${pool_cfg.name || 'unnamed'}) PG ${pg_num}`+
|
|
|
|
` primary OSD from ${pg_cfg.primary} to ${new_primary}`
|
|
|
|
);
|
|
|
|
changed = true;
|
|
|
|
pg_cfg.primary = new_primary;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
if (changed)
|
|
|
|
{
|
|
|
|
await this.save_pg_config();
|
2020-05-14 19:29:35 +03:00
|
|
|
}
|
2020-05-12 17:49:18 +03:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2020-09-13 12:23:11 +03:00
|
|
|
async save_pg_config(etcd_request = { compare: [], success: [] })
|
|
|
|
{
|
|
|
|
etcd_request.compare.push(
|
|
|
|
{ key: b64(this.etcd_prefix+'/mon/master'), target: 'LEASE', lease: ''+this.etcd_lease_id },
|
|
|
|
{ key: b64(this.etcd_prefix+'/config/pgs'), target: 'MOD', mod_revision: ''+this.etcd_watch_revision, result: 'LESS' },
|
|
|
|
);
|
|
|
|
etcd_request.success.push(
|
|
|
|
{ requestPut: { key: b64(this.etcd_prefix+'/config/pgs'), value: b64(JSON.stringify(this.state.config.pgs)) } },
|
|
|
|
);
|
|
|
|
const res = await this.etcd_call('/kv/txn', etcd_request, this.config.etcd_mon_timeout, 0);
|
|
|
|
if (!res.succeeded)
|
|
|
|
{
|
|
|
|
console.log('Someone changed PG configuration while we also tried to change it. Retrying in '+this.config.mon_change_timeout+' ms');
|
|
|
|
this.schedule_recheck();
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
console.log('PG configuration successfully changed');
|
|
|
|
}
|
|
|
|
|
|
|
|
// Schedule next recheck at least at <unixtime>
|
|
|
|
schedule_next_recheck_at(unixtime)
|
|
|
|
{
|
|
|
|
this.next_recheck_at = !this.next_recheck_at || this.next_recheck_at > unixtime
|
|
|
|
? unixtime : this.next_recheck_at;
|
|
|
|
const now = Date.now()/1000;
|
|
|
|
if (this.next_recheck_timer)
|
|
|
|
{
|
|
|
|
clearTimeout(this.next_recheck_timer);
|
|
|
|
this.next_recheck_timer = null;
|
|
|
|
}
|
|
|
|
if (this.next_recheck_at < now)
|
|
|
|
{
|
|
|
|
this.next_recheck_at = 0;
|
|
|
|
this.schedule_recheck();
|
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
|
|
|
this.next_recheck_timer = setTimeout(() =>
|
|
|
|
{
|
|
|
|
this.next_recheck_timer = null;
|
|
|
|
this.next_recheck_at = 0;
|
|
|
|
this.schedule_recheck();
|
|
|
|
}, now-this.next_recheck_at);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
// Schedule a recheck to run after a small timeout (1s)
|
|
|
|
// If already scheduled, cancel previous timer and schedule it again
|
|
|
|
// This is required for multiple change events to trigger at most 1 recheck in 1s
|
2020-05-14 19:29:35 +03:00
|
|
|
schedule_recheck()
|
|
|
|
{
|
2020-05-15 01:28:44 +03:00
|
|
|
if (this.recheck_timer)
|
2020-05-14 19:29:35 +03:00
|
|
|
{
|
2020-05-15 01:28:44 +03:00
|
|
|
clearTimeout(this.recheck_timer);
|
|
|
|
this.recheck_timer = null;
|
2020-05-14 19:29:35 +03:00
|
|
|
}
|
2020-05-15 01:28:44 +03:00
|
|
|
this.recheck_timer = setTimeout(() =>
|
2020-05-14 19:29:35 +03:00
|
|
|
{
|
2020-05-15 01:28:44 +03:00
|
|
|
this.recheck_timer = null;
|
2021-11-28 01:19:42 +03:00
|
|
|
this.recheck_pgs().catch(this.die);
|
2020-05-14 19:29:35 +03:00
|
|
|
}, this.config.mon_change_timeout || 1000);
|
|
|
|
}
|
|
|
|
|
2021-11-10 23:51:15 +03:00
|
|
|
sum_op_stats(timestamp, prev_stats)
|
2020-05-15 01:28:44 +03:00
|
|
|
{
|
|
|
|
const op_stats = {}, subop_stats = {}, recovery_stats = {};
|
|
|
|
for (const osd in this.state.osd.stats)
|
|
|
|
{
|
2021-03-27 01:15:05 +03:00
|
|
|
const st = this.state.osd.stats[osd]||{};
|
2020-05-15 01:28:44 +03:00
|
|
|
for (const op in st.op_stats||{})
|
|
|
|
{
|
|
|
|
op_stats[op] = op_stats[op] || { count: 0n, usec: 0n, bytes: 0n };
|
2021-02-28 19:51:14 +03:00
|
|
|
op_stats[op].count += BigInt(st.op_stats[op].count||0);
|
|
|
|
op_stats[op].usec += BigInt(st.op_stats[op].usec||0);
|
|
|
|
op_stats[op].bytes += BigInt(st.op_stats[op].bytes||0);
|
2020-05-15 01:28:44 +03:00
|
|
|
}
|
|
|
|
for (const op in st.subop_stats||{})
|
|
|
|
{
|
|
|
|
subop_stats[op] = subop_stats[op] || { count: 0n, usec: 0n };
|
2021-02-28 19:51:14 +03:00
|
|
|
subop_stats[op].count += BigInt(st.subop_stats[op].count||0);
|
|
|
|
subop_stats[op].usec += BigInt(st.subop_stats[op].usec||0);
|
2020-05-15 01:28:44 +03:00
|
|
|
}
|
|
|
|
for (const op in st.recovery_stats||{})
|
|
|
|
{
|
|
|
|
recovery_stats[op] = recovery_stats[op] || { count: 0n, bytes: 0n };
|
2021-02-28 19:51:14 +03:00
|
|
|
recovery_stats[op].count += BigInt(st.recovery_stats[op].count||0);
|
|
|
|
recovery_stats[op].bytes += BigInt(st.recovery_stats[op].bytes||0);
|
2020-05-15 01:28:44 +03:00
|
|
|
}
|
|
|
|
}
|
2021-11-10 23:51:15 +03:00
|
|
|
if (prev_stats && prev_stats.timestamp >= timestamp)
|
|
|
|
{
|
|
|
|
prev_stats = null;
|
|
|
|
}
|
|
|
|
const tm = prev_stats ? BigInt(timestamp - prev_stats.timestamp) : 0;
|
|
|
|
for (const op in op_stats)
|
|
|
|
{
|
|
|
|
op_stats[op].bps = prev_stats ? (op_stats[op].bytes - prev_stats.op_stats[op].bytes) * 1000n / tm : 0;
|
|
|
|
op_stats[op].iops = prev_stats ? (op_stats[op].count - prev_stats.op_stats[op].count) * 1000n / tm : 0;
|
|
|
|
op_stats[op].lat = prev_stats ? (op_stats[op].usec - prev_stats.op_stats[op].usec)
|
|
|
|
/ ((op_stats[op].count - prev_stats.op_stats[op].count) || 1n) : 0;
|
|
|
|
}
|
|
|
|
for (const op in subop_stats)
|
|
|
|
{
|
|
|
|
subop_stats[op].iops = prev_stats ? (subop_stats[op].count - prev_stats.subop_stats[op].count) * 1000n / tm : 0;
|
|
|
|
subop_stats[op].lat = prev_stats ? (subop_stats[op].usec - prev_stats.subop_stats[op].usec)
|
|
|
|
/ ((subop_stats[op].count - prev_stats.subop_stats[op].count) || 1n) : 0;
|
|
|
|
}
|
|
|
|
for (const op in recovery_stats)
|
|
|
|
{
|
|
|
|
recovery_stats[op].bps = prev_stats ? (recovery_stats[op].bytes - prev_stats.recovery_stats[op].bytes) * 1000n / tm : 0;
|
|
|
|
recovery_stats[op].iops = prev_stats ? (recovery_stats[op].count - prev_stats.recovery_stats[op].count) * 1000n / tm : 0;
|
|
|
|
}
|
2021-01-21 00:30:18 +03:00
|
|
|
return { op_stats, subop_stats, recovery_stats };
|
|
|
|
}
|
|
|
|
|
|
|
|
sum_object_counts()
|
|
|
|
{
|
2020-05-15 01:28:44 +03:00
|
|
|
const object_counts = { object: 0n, clean: 0n, misplaced: 0n, degraded: 0n, incomplete: 0n };
|
2020-09-01 18:50:23 +03:00
|
|
|
for (const pool_id in this.state.pg.stats)
|
2020-05-15 01:28:44 +03:00
|
|
|
{
|
2020-09-01 18:50:23 +03:00
|
|
|
for (const pg_num in this.state.pg.stats[pool_id])
|
2020-05-15 01:28:44 +03:00
|
|
|
{
|
2020-09-01 18:50:23 +03:00
|
|
|
const st = this.state.pg.stats[pool_id][pg_num];
|
2021-03-08 23:13:08 +03:00
|
|
|
if (st)
|
2020-05-15 01:28:44 +03:00
|
|
|
{
|
2021-03-08 23:13:08 +03:00
|
|
|
for (const k in object_counts)
|
2020-09-01 18:50:23 +03:00
|
|
|
{
|
2021-03-08 23:13:08 +03:00
|
|
|
if (st[k+'_count'])
|
|
|
|
{
|
|
|
|
object_counts[k] += BigInt(st[k+'_count']);
|
|
|
|
}
|
2020-09-01 18:50:23 +03:00
|
|
|
}
|
2020-05-15 01:28:44 +03:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
2021-01-21 00:30:18 +03:00
|
|
|
return object_counts;
|
2020-05-15 01:28:44 +03:00
|
|
|
}
|
|
|
|
|
2021-11-10 23:51:15 +03:00
|
|
|
sum_inode_stats(prev_stats, timestamp, prev_timestamp)
|
2020-05-15 01:28:44 +03:00
|
|
|
{
|
2021-01-21 00:30:18 +03:00
|
|
|
const inode_stats = {};
|
|
|
|
const inode_stub = () => ({
|
|
|
|
raw_used: 0n,
|
|
|
|
read: { count: 0n, usec: 0n, bytes: 0n },
|
|
|
|
write: { count: 0n, usec: 0n, bytes: 0n },
|
|
|
|
delete: { count: 0n, usec: 0n, bytes: 0n },
|
|
|
|
});
|
2021-10-21 14:00:54 +03:00
|
|
|
const seen_pools = {};
|
2021-07-02 22:47:01 +03:00
|
|
|
for (const pool_id in this.state.config.pools)
|
|
|
|
{
|
2021-10-21 14:00:54 +03:00
|
|
|
seen_pools[pool_id] = true;
|
2021-07-02 22:47:01 +03:00
|
|
|
this.state.pool.stats[pool_id] = this.state.pool.stats[pool_id] || {};
|
|
|
|
this.state.pool.stats[pool_id].used_raw_tb = 0n;
|
|
|
|
}
|
2021-01-21 00:30:18 +03:00
|
|
|
for (const osd_num in this.state.osd.space)
|
|
|
|
{
|
2021-04-04 11:52:11 +03:00
|
|
|
for (const pool_id in this.state.osd.space[osd_num])
|
2021-01-21 00:30:18 +03:00
|
|
|
{
|
2021-10-21 14:00:54 +03:00
|
|
|
this.state.pool.stats[pool_id] = this.state.pool.stats[pool_id] || {};
|
|
|
|
if (!seen_pools[pool_id])
|
|
|
|
{
|
|
|
|
this.state.pool.stats[pool_id].used_raw_tb = 0n;
|
|
|
|
seen_pools[pool_id] = true;
|
|
|
|
}
|
2021-04-04 11:52:11 +03:00
|
|
|
inode_stats[pool_id] = inode_stats[pool_id] || {};
|
|
|
|
for (const inode_num in this.state.osd.space[osd_num][pool_id])
|
|
|
|
{
|
2021-07-02 22:47:01 +03:00
|
|
|
const u = BigInt(this.state.osd.space[osd_num][pool_id][inode_num]||0);
|
2021-04-04 11:52:11 +03:00
|
|
|
inode_stats[pool_id][inode_num] = inode_stats[pool_id][inode_num] || inode_stub();
|
2021-07-02 22:47:01 +03:00
|
|
|
inode_stats[pool_id][inode_num].raw_used += u;
|
|
|
|
this.state.pool.stats[pool_id].used_raw_tb += u;
|
2021-04-04 11:52:11 +03:00
|
|
|
}
|
2021-01-21 00:30:18 +03:00
|
|
|
}
|
|
|
|
}
|
2021-10-21 14:00:54 +03:00
|
|
|
for (const pool_id in seen_pools)
|
2021-07-02 22:47:01 +03:00
|
|
|
{
|
|
|
|
const used = this.state.pool.stats[pool_id].used_raw_tb;
|
|
|
|
this.state.pool.stats[pool_id].used_raw_tb = Number(used)/1024/1024/1024/1024;
|
|
|
|
}
|
2021-01-21 00:30:18 +03:00
|
|
|
for (const osd_num in this.state.osd.inodestats)
|
2020-05-15 01:28:44 +03:00
|
|
|
{
|
2021-01-21 00:30:18 +03:00
|
|
|
const ist = this.state.osd.inodestats[osd_num];
|
2021-04-04 11:52:11 +03:00
|
|
|
for (const pool_id in ist)
|
2020-05-15 01:28:44 +03:00
|
|
|
{
|
2021-04-04 11:52:11 +03:00
|
|
|
inode_stats[pool_id] = inode_stats[pool_id] || {};
|
|
|
|
for (const inode_num in ist[pool_id])
|
2020-05-15 01:28:44 +03:00
|
|
|
{
|
2021-04-04 11:52:11 +03:00
|
|
|
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);
|
|
|
|
}
|
2021-01-21 00:30:18 +03:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
2021-11-10 23:51:15 +03:00
|
|
|
if (prev_stats && prev_timestamp >= timestamp)
|
2021-01-21 00:30:18 +03:00
|
|
|
{
|
2021-11-10 23:51:15 +03:00
|
|
|
prev_stats = null;
|
|
|
|
}
|
|
|
|
const tm = prev_stats ? BigInt(timestamp - prev_timestamp) : 0;
|
|
|
|
for (const pool_id in inode_stats)
|
|
|
|
{
|
|
|
|
for (const inode_num in inode_stats[pool_id])
|
2021-01-21 00:30:18 +03:00
|
|
|
{
|
2021-11-28 21:02:05 +03:00
|
|
|
let nonzero = inode_stats[pool_id][inode_num].raw_used > 0;
|
2021-11-10 23:51:15 +03:00
|
|
|
for (const op of [ 'read', 'write', 'delete' ])
|
2021-01-21 00:30:18 +03:00
|
|
|
{
|
2021-11-10 23:51:15 +03:00
|
|
|
const op_st = inode_stats[pool_id][inode_num][op];
|
|
|
|
const prev_st = prev_stats && prev_stats[pool_id] && prev_stats[pool_id][inode_num] && prev_stats[pool_id][inode_num][op];
|
|
|
|
op_st.bps = prev_st ? (op_st.bytes - prev_st.bytes) * 1000n / tm : 0;
|
|
|
|
op_st.iops = prev_st ? (op_st.count - prev_st.count) * 1000n / tm : 0;
|
|
|
|
op_st.lat = prev_st ? (op_st.usec - prev_st.usec) / ((op_st.count - prev_st.count) || 1n) : 0;
|
2021-11-28 21:02:05 +03:00
|
|
|
if (op_st.bps > 0 || op_st.iops > 0 || op_st.lat > 0)
|
|
|
|
nonzero = true;
|
|
|
|
}
|
|
|
|
if (!nonzero && (!this.state.config.inode[pool_id] || !this.state.config.inode[pool_id][inode_num]))
|
|
|
|
{
|
|
|
|
// Deleted inode (no data, no I/O, no config)
|
|
|
|
delete inode_stats[pool_id][inode_num];
|
2020-05-15 01:28:44 +03:00
|
|
|
}
|
|
|
|
}
|
2021-01-19 02:14:13 +03:00
|
|
|
}
|
2021-11-10 23:51:15 +03:00
|
|
|
return inode_stats;
|
2021-01-21 00:30:18 +03:00
|
|
|
}
|
|
|
|
|
|
|
|
serialize_bigints(obj)
|
|
|
|
{
|
2021-11-10 23:51:15 +03:00
|
|
|
obj = { ...obj };
|
2021-01-21 00:30:18 +03:00
|
|
|
for (const k in obj)
|
2021-01-19 02:14:13 +03:00
|
|
|
{
|
2021-01-21 00:30:18 +03:00
|
|
|
if (typeof obj[k] == 'bigint')
|
|
|
|
{
|
|
|
|
obj[k] = ''+obj[k];
|
|
|
|
}
|
|
|
|
else if (typeof obj[k] == 'object')
|
2021-01-19 02:14:13 +03:00
|
|
|
{
|
2021-11-10 23:51:15 +03:00
|
|
|
obj[k] = this.serialize_bigints(obj[k]);
|
2021-01-19 02:14:13 +03:00
|
|
|
}
|
|
|
|
}
|
2021-11-10 23:51:15 +03:00
|
|
|
return obj;
|
2021-01-21 00:30:18 +03:00
|
|
|
}
|
|
|
|
|
|
|
|
async update_total_stats()
|
|
|
|
{
|
|
|
|
const txn = [];
|
2021-11-10 23:51:15 +03:00
|
|
|
const timestamp = Date.now();
|
2021-01-21 00:30:18 +03:00
|
|
|
const object_counts = this.sum_object_counts();
|
2021-11-10 23:51:15 +03:00
|
|
|
let stats = this.sum_op_stats(timestamp, this.prev_stats);
|
|
|
|
let inode_stats = this.sum_inode_stats(
|
|
|
|
this.prev_stats ? this.prev_stats.inode_stats : null,
|
|
|
|
timestamp, this.prev_stats ? this.prev_stats.timestamp : null
|
|
|
|
);
|
|
|
|
this.prev_stats = { timestamp, ...stats, inode_stats };
|
2021-01-21 00:30:18 +03:00
|
|
|
stats.object_counts = object_counts;
|
2021-11-10 23:51:15 +03:00
|
|
|
stats = this.serialize_bigints(stats);
|
|
|
|
inode_stats = this.serialize_bigints(inode_stats);
|
2021-01-21 00:30:18 +03:00
|
|
|
txn.push({ requestPut: { key: b64(this.etcd_prefix+'/stats'), value: b64(JSON.stringify(stats)) } });
|
2021-04-04 11:52:11 +03:00
|
|
|
for (const pool_id in inode_stats)
|
2021-01-19 02:14:13 +03:00
|
|
|
{
|
2021-04-04 11:52:11 +03:00
|
|
|
for (const inode_num in inode_stats[pool_id])
|
|
|
|
{
|
|
|
|
txn.push({ requestPut: {
|
|
|
|
key: b64(this.etcd_prefix+'/inode/stats/'+pool_id+'/'+inode_num),
|
|
|
|
value: b64(JSON.stringify(inode_stats[pool_id][inode_num])),
|
|
|
|
} });
|
|
|
|
}
|
2021-01-19 02:14:13 +03:00
|
|
|
}
|
2021-11-28 21:02:05 +03:00
|
|
|
for (const pool_id in this.state.inode.stats)
|
|
|
|
{
|
|
|
|
for (const inode_num in this.state.inode.stats[pool_id])
|
|
|
|
{
|
|
|
|
if (!inode_stats[pool_id] || !inode_stats[pool_id][inode_num])
|
|
|
|
{
|
|
|
|
txn.push({ requestDeleteRange: {
|
|
|
|
key: b64(this.etcd_prefix+'/inode/stats/'+pool_id+'/'+inode_num),
|
|
|
|
} });
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
2021-07-02 22:47:01 +03:00
|
|
|
for (const pool_id in this.state.pool.stats)
|
|
|
|
{
|
2021-10-21 14:00:54 +03:00
|
|
|
const pool_stats = { ...this.state.pool.stats[pool_id] };
|
|
|
|
this.serialize_bigints(pool_stats);
|
2021-07-02 22:47:01 +03:00
|
|
|
txn.push({ requestPut: {
|
|
|
|
key: b64(this.etcd_prefix+'/pool/stats/'+pool_id),
|
2021-10-21 14:00:54 +03:00
|
|
|
value: b64(JSON.stringify(pool_stats)),
|
2021-07-02 22:47:01 +03:00
|
|
|
} });
|
|
|
|
}
|
2021-01-19 02:14:13 +03:00
|
|
|
if (txn.length)
|
|
|
|
{
|
|
|
|
await this.etcd_call('/kv/txn', { success: txn }, this.config.etcd_mon_timeout, 0);
|
2020-05-15 01:28:44 +03:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
schedule_update_stats()
|
|
|
|
{
|
|
|
|
if (this.stats_timer)
|
|
|
|
{
|
2021-11-09 01:41:22 +03:00
|
|
|
return;
|
2021-07-02 22:47:01 +03:00
|
|
|
}
|
2020-05-15 01:28:44 +03:00
|
|
|
this.stats_timer = setTimeout(() =>
|
|
|
|
{
|
|
|
|
this.stats_timer = null;
|
|
|
|
this.update_total_stats().catch(console.error);
|
2021-11-09 01:41:22 +03:00
|
|
|
}, this.config.mon_stats_timeout);
|
2020-05-15 01:28:44 +03:00
|
|
|
}
|
|
|
|
|
2020-05-12 17:49:18 +03:00
|
|
|
parse_kv(kv)
|
|
|
|
{
|
|
|
|
if (!kv || !kv.key)
|
|
|
|
{
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
kv.key = de64(kv.key);
|
2020-09-05 02:14:43 +03:00
|
|
|
kv.value = kv.value ? de64(kv.value) : null;
|
|
|
|
let key = kv.key.substr(this.etcd_prefix.length+1);
|
2020-09-01 17:07:06 +03:00
|
|
|
if (!this.constructor.etcd_allow.exec(key))
|
2020-05-12 17:49:18 +03:00
|
|
|
{
|
|
|
|
console.log('Bad key in etcd: '+kv.key+' = '+kv.value);
|
|
|
|
return;
|
|
|
|
}
|
2020-09-11 19:25:52 +03:00
|
|
|
try
|
|
|
|
{
|
|
|
|
kv.value = kv.value ? JSON.parse(kv.value) : null;
|
|
|
|
}
|
|
|
|
catch (e)
|
|
|
|
{
|
2020-10-23 01:54:38 +03:00
|
|
|
console.log('Bad value in etcd: '+kv.key+' = '+kv.value);
|
2020-09-11 19:25:52 +03:00
|
|
|
return;
|
|
|
|
}
|
2020-03-12 00:56:41 +03:00
|
|
|
let key_parts = key.split('/');
|
2020-09-01 18:50:23 +03:00
|
|
|
let cur = this.state;
|
2020-03-12 00:56:41 +03:00
|
|
|
for (let i = 0; i < key_parts.length-1; i++)
|
|
|
|
{
|
|
|
|
cur = (cur[key_parts[i]] = cur[key_parts[i]] || {});
|
|
|
|
}
|
|
|
|
if (etcd_nonempty_keys[key])
|
2020-09-01 17:07:06 +03:00
|
|
|
{
|
2020-03-12 00:56:41 +03:00
|
|
|
// Do not clear these to null
|
|
|
|
kv.value = kv.value || {};
|
2020-09-01 17:07:06 +03:00
|
|
|
}
|
2020-03-12 00:56:41 +03:00
|
|
|
cur[key_parts[key_parts.length-1]] = kv.value;
|
|
|
|
if (key === 'config/global')
|
2020-05-12 17:49:18 +03:00
|
|
|
{
|
2021-11-28 01:19:42 +03:00
|
|
|
this.config = { ...this.initConfig, ...this.state.config.global };
|
2020-05-14 19:29:35 +03:00
|
|
|
this.check_config();
|
2020-09-13 12:23:11 +03:00
|
|
|
for (const osd_num in this.state.osd.stats)
|
|
|
|
{
|
|
|
|
// Recheck PGs <osd_out_time> later
|
|
|
|
this.schedule_next_recheck_at(
|
|
|
|
!this.state.osd.stats[osd_num] ? 0 : this.state.osd.stats[osd_num].time+this.config.osd_out_time
|
|
|
|
);
|
|
|
|
}
|
2020-05-12 17:49:18 +03:00
|
|
|
}
|
2020-03-12 00:56:41 +03:00
|
|
|
else if (key === 'config/pools')
|
2020-09-05 02:14:43 +03:00
|
|
|
{
|
|
|
|
for (const pool_id in this.state.config.pools)
|
|
|
|
{
|
2020-09-13 12:23:11 +03:00
|
|
|
// Adjust pool configuration so PG distribution hash doesn't change on recheck()
|
2020-09-05 02:14:43 +03:00
|
|
|
const pool_cfg = this.state.config.pools[pool_id];
|
|
|
|
this.validate_pool_cfg(pool_id, pool_cfg, true);
|
|
|
|
}
|
|
|
|
}
|
2020-03-12 00:56:41 +03:00
|
|
|
else if (key_parts[0] === 'osd' && key_parts[1] === 'stats')
|
2020-09-13 12:23:11 +03:00
|
|
|
{
|
|
|
|
// Recheck PGs <osd_out_time> later
|
|
|
|
this.schedule_next_recheck_at(
|
|
|
|
!this.state.osd.stats[key[2]] ? 0 : this.state.osd.stats[key[2]].time+this.config.osd_out_time
|
|
|
|
);
|
|
|
|
}
|
2020-05-12 17:49:18 +03:00
|
|
|
}
|
|
|
|
|
|
|
|
async etcd_call(path, body, timeout, retries)
|
|
|
|
{
|
|
|
|
let retry = 0;
|
|
|
|
if (retries >= 0 && retries < 1)
|
|
|
|
{
|
|
|
|
retries = 1;
|
|
|
|
}
|
|
|
|
while (retries < 0 || retry < retries)
|
|
|
|
{
|
2021-11-28 01:19:42 +03:00
|
|
|
retry++;
|
|
|
|
const base = this.pick_next_etcd();
|
2020-05-12 17:49:18 +03:00
|
|
|
const res = await POST(base+path, body, timeout);
|
2020-08-03 23:50:50 +03:00
|
|
|
if (res.error)
|
|
|
|
{
|
2021-11-28 01:19:42 +03:00
|
|
|
if (this.selected_etcd_url == base)
|
|
|
|
this.selected_etcd_url = null;
|
|
|
|
console.error('failed to query etcd: '+res.error);
|
|
|
|
continue;
|
2020-08-03 23:50:50 +03:00
|
|
|
}
|
2020-05-12 17:49:18 +03:00
|
|
|
if (res.json)
|
|
|
|
{
|
|
|
|
if (res.json.error)
|
|
|
|
{
|
2020-09-05 02:14:43 +03:00
|
|
|
console.error('etcd returned error: '+res.json.error);
|
2020-05-12 17:49:18 +03:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
return res.json;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
this.die();
|
|
|
|
}
|
|
|
|
|
2021-11-28 01:19:42 +03:00
|
|
|
_die(err)
|
2020-05-12 17:49:18 +03:00
|
|
|
{
|
|
|
|
// In fact we can just try to rejoin
|
2020-09-05 02:14:43 +03:00
|
|
|
console.error(new Error(err || 'Cluster connection failed'));
|
2020-05-12 17:49:18 +03:00
|
|
|
process.exit(1);
|
|
|
|
}
|
|
|
|
|
2021-11-28 01:19:42 +03:00
|
|
|
local_ips(all)
|
2020-05-12 17:49:18 +03:00
|
|
|
{
|
|
|
|
const ips = [];
|
|
|
|
const ifaces = os.networkInterfaces();
|
|
|
|
for (const ifname in ifaces)
|
|
|
|
{
|
|
|
|
for (const iface of ifaces[ifname])
|
|
|
|
{
|
2021-11-28 01:19:42 +03:00
|
|
|
if (iface.family == 'IPv4' && !iface.internal || all)
|
2020-05-12 17:49:18 +03:00
|
|
|
{
|
|
|
|
ips.push(iface.address);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return ips;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
function POST(url, body, timeout)
|
|
|
|
{
|
|
|
|
return new Promise((ok, no) =>
|
|
|
|
{
|
|
|
|
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',
|
2020-08-03 23:50:50 +03:00
|
|
|
'Content-Length': body_text.length,
|
2020-05-12 17:49:18 +03:00
|
|
|
} }, (res) =>
|
|
|
|
{
|
|
|
|
if (!req)
|
|
|
|
{
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
clearTimeout(timer_id);
|
|
|
|
let res_body = '';
|
|
|
|
res.setEncoding('utf8');
|
2020-09-01 18:50:23 +03:00
|
|
|
res.on('data', chunk => { res_body += chunk; });
|
2020-05-12 17:49:18 +03:00
|
|
|
res.on('end', () =>
|
|
|
|
{
|
2020-08-03 23:50:50 +03:00
|
|
|
if (res.statusCode != 200)
|
|
|
|
{
|
|
|
|
ok({ error: res_body, code: res.statusCode });
|
|
|
|
return;
|
|
|
|
}
|
2020-05-12 17:49:18 +03:00
|
|
|
try
|
|
|
|
{
|
|
|
|
res_body = JSON.parse(res_body);
|
|
|
|
ok({ response: res, json: res_body });
|
|
|
|
}
|
|
|
|
catch (e)
|
|
|
|
{
|
|
|
|
ok({ error: e, response: res, body: res_body });
|
|
|
|
}
|
|
|
|
});
|
|
|
|
});
|
|
|
|
req.write(body_text);
|
|
|
|
req.end();
|
|
|
|
});
|
|
|
|
}
|
|
|
|
|
|
|
|
function b64(str)
|
|
|
|
{
|
|
|
|
return Buffer.from(str).toString('base64');
|
|
|
|
}
|
|
|
|
|
|
|
|
function de64(str)
|
|
|
|
{
|
|
|
|
return Buffer.from(str, 'base64').toString();
|
|
|
|
}
|
|
|
|
|
|
|
|
function sha1hex(str)
|
|
|
|
{
|
|
|
|
const hash = crypto.createHash('sha1');
|
|
|
|
hash.update(str);
|
|
|
|
return hash.digest('hex');
|
|
|
|
}
|
2020-09-05 02:14:43 +03:00
|
|
|
|
2020-11-08 01:54:12 +03:00
|
|
|
Mon.etcd_allow = etcd_allow;
|
|
|
|
Mon.etcd_tree = etcd_tree;
|
|
|
|
|
2020-09-05 02:14:43 +03:00
|
|
|
module.exports = Mon;
|