Compare commits

...

34 Commits

Author SHA1 Message Date
Vitaliy Filippov 06f4e0fcce K/V control prints (for debug only) O:-)
Test / test_move_reappear (push) Successful in 18s Details
Test / test_rm (push) Successful in 12s Details
Test / test_snapshot_chain (push) Successful in 59s Details
Test / test_snapshot_down (push) Successful in 19s Details
Test / test_snapshot_ec (push) Failing after 3m5s Details
Test / test_splitbrain (push) Successful in 12s Details
Test / test_snapshot_chain_ec (push) Failing after 3m6s Details
Test / test_snapshot_down_ec (push) Failing after 3m10s Details
Test / test_rebalance_verify_ec (push) Failing after 43s Details
Test / test_rebalance_verify_imm (push) Successful in 3m0s Details
Test / test_rebalance_verify (push) Successful in 3m29s Details
Test / test_rebalance_verify_ec_imm (push) Successful in 2m48s Details
Test / test_write_no_same (push) Successful in 12s Details
Test / test_interrupted_rebalance_ec (push) Failing after 10m5s Details
Test / test_write (push) Failing after 3m6s Details
Test / test_write_xor (push) Failing after 3m5s Details
Test / test_heal_pg_size_2 (push) Failing after 3m46s Details
Test / test_heal_csum_32k_dj (push) Successful in 4m48s Details
Test / test_heal_csum_32k_dmj (push) Failing after 4m50s Details
Test / test_heal_csum_32k (push) Successful in 5m9s Details
Test / test_heal_ec (push) Failing after 10m14s Details
Test / test_heal_csum_4k_dj (push) Successful in 5m28s Details
Test / test_scrub (push) Successful in 1m11s Details
Test / test_heal_csum_4k_dmj (push) Failing after 5m36s Details
Test / test_scrub_zero_osd_2 (push) Successful in 45s Details
Test / test_scrub_pg_size_3 (push) Successful in 1m2s Details
Test / test_scrub_xor (push) Failing after 3m8s Details
Test / test_scrub_pg_size_6_pg_minsize_4_osd_count_6_ec (push) Failing after 3m8s Details
Test / test_scrub_ec (push) Failing after 3m6s Details
Test / test_heal_csum_4k (push) Failing after 10m19s Details
2023-12-01 02:33:04 +03:00
Vitaliy Filippov f285cfc483 Fix eviction when random_pos selects the end
Test / test_move_reappear (push) Successful in 18s Details
Test / test_rm (push) Successful in 13s Details
Test / test_snapshot_chain (push) Successful in 1m1s Details
Test / test_snapshot_down (push) Successful in 20s Details
Test / test_snapshot_ec (push) Failing after 3m5s Details
Test / test_splitbrain (push) Successful in 12s Details
Test / test_snapshot_chain_ec (push) Failing after 3m6s Details
Test / test_snapshot_down_ec (push) Failing after 3m6s Details
Test / test_rebalance_verify_ec (push) Failing after 45s Details
Test / test_rebalance_verify (push) Successful in 2m34s Details
Test / test_rebalance_verify_imm (push) Successful in 2m5s Details
Test / test_write (push) Successful in 54s Details
Test / test_write_no_same (push) Successful in 12s Details
Test / test_rebalance_verify_ec_imm (push) Successful in 3m7s Details
Test / test_write_xor (push) Failing after 3m6s Details
Test / test_interrupted_rebalance_ec (push) Failing after 10m6s Details
Test / test_heal_pg_size_2 (push) Failing after 10m10s Details
Test / test_heal_ec (push) Failing after 10m7s Details
Test / test_heal_csum_32k_dmj (push) Failing after 10m10s Details
Test / test_heal_csum_32k_dj (push) Failing after 10m15s Details
Test / test_heal_csum_32k (push) Failing after 3m29s Details
Test / test_scrub (push) Successful in 1m9s Details
Test / test_scrub_zero_osd_2 (push) Successful in 1m18s Details
Test / test_heal_csum_4k_dmj (push) Failing after 5m56s Details
Test / test_scrub_pg_size_3 (push) Successful in 31s Details
Test / test_scrub_xor (push) Failing after 3m12s Details
Test / test_scrub_pg_size_6_pg_minsize_4_osd_count_6_ec (push) Failing after 3m6s Details
Test / test_heal_csum_4k_dj (push) Failing after 10m10s Details
Test / test_scrub_ec (push) Failing after 3m6s Details
Test / test_heal_csum_4k (push) Failing after 10m21s Details
2023-12-01 01:43:03 +03:00
Vitaliy Filippov 12b50b421d Implement min/max list_count to make listings during performance test reasonable 2023-12-01 01:17:04 +03:00
Vitaliy Filippov 9f6d09428d Fix and improve parallel allocation
- Do not try to allocate more DB blocks in an inode block until it's "confirmed" and "locked" by the first write
- Do not recheck for new zero DB blocks on first write into an inode block - a CAS failure means someone else is already writing into it
- Throw new allocation blocks away regardless of whether the known_version is 0 on a CAS failure
2023-12-01 01:17:04 +03:00
Vitaliy Filippov 580025cfc9 Implement key_prefix for K/V stress test 2023-12-01 01:17:04 +03:00
Vitaliy Filippov 13e2d3ce7c More fixes
- do not overwrite a block with older version if known version is newer
  (read may start before update and end after update)
- invalidated block versions can't be remembered and trusted
- right boundary for split blocks is right_half when diving down, not key_lt
- restart update also when block is "invalidated", not just on version mismatch
- copy callback in listings to avoid closure destruction bugs too
2023-12-01 01:17:04 +03:00
Vitaliy Filippov c5b00f897a Add logging and one more assert 2023-12-01 01:17:04 +03:00
Vitaliy Filippov e847e26912 Make get_block() wait for updating when unrelated block is found along the path 2023-12-01 01:17:04 +03:00
Vitaliy Filippov 3393463466 Fix a race condition where changed blocks were parsed over existing cached blocks and getting a mix of data 2023-12-01 01:17:04 +03:00
Vitaliy Filippov bd96a6194a Simplify code by removing an unneeded "optimisation" 2023-12-01 01:17:04 +03:00
Vitaliy Filippov 601fe10c28 Add kv_log_level, print warnings on level 1, trace ops on level 10 2023-12-01 01:17:04 +03:00
Vitaliy Filippov 63dbc9ca85 Fix duplicate keys in listings on parallel updates -- do not rewind key "iterator position" 2023-12-01 01:17:04 +03:00
Vitaliy Filippov aa0c363c39 Implement key suffix to avoid collisions of multiple test workers 2023-12-01 01:17:04 +03:00
Vitaliy Filippov ce52c5589e Do not complain on empty first block 2023-12-01 01:17:04 +03:00
Vitaliy Filippov aee20ab1ee Add JSON output for stress-tester 2023-12-01 01:17:04 +03:00
Vitaliy Filippov bb81992fac Print total stats 2023-12-01 01:17:04 +03:00
Vitaliy Filippov a28f401aff Do not send more than op_count operations (fix segfault on finish) 2023-12-01 01:17:04 +03:00
Vitaliy Filippov 4ac7e096fd Add some more resiliency to serialize() 2023-12-01 01:17:04 +03:00
Vitaliy Filippov b6171a4599 Invalidate blocks being updated too 2023-12-01 01:17:03 +03:00
Vitaliy Filippov 28045f230c Change new block allocation method: make each writer choose multiple empty PG blocks and place blocks in them 2023-12-01 01:17:03 +03:00
Vitaliy Filippov 10e867880f Remove blocks from cache on unsuccessful updates 2023-12-01 01:17:03 +03:00
Vitaliy Filippov 012462171a Allow to track multiple updates per block (it should never happen though) 2023-12-01 01:17:03 +03:00
Vitaliy Filippov 904793cdab Do not call stop_updating after failed write_new_block and after clear_block (both delete the item) 2023-12-01 01:17:03 +03:00
Vitaliy Filippov 45c01db2de Track versions of parent blocks and recheck if changed during update 2023-12-01 01:17:03 +03:00
Vitaliy Filippov 8c9206cecd Fix resume_split condition (key_lt can also be "") 2023-12-01 01:17:03 +03:00
Vitaliy Filippov e8c46ededa Experiment: transform offsets for better sharding 2023-12-01 01:17:03 +03:00
Vitaliy Filippov e9b321a0e0 More post-stress-test fixes
- Prevent _split types of new blocks
- Stop updating new blocks only after the whole update, otherwise pointers
  may become invalid
- Use recheck_none for updates initially
- Use UINT64_MAX as initial block version when postponing ops, otherwise the
  check fails when the block is initially empty. This for example leads to
  writing both leaf items & block pointers (which is incorrect) into the root
  block when starting stress-test with --parallelism 32
- Fix -EINTR comparison
2023-12-01 01:17:03 +03:00
Vitaliy Filippov 09a77991ae Print operation statistics 2023-12-01 01:17:03 +03:00
Vitaliy Filippov 29d8c9b6f3 K/V fixes after stress-test :-)
- track block versions correctly - per inode block (128kb) instead of tree block (4kb)
- prevent multiple parallel CAS writes of the same inode block
- add logging for EILSEQ which means invalid data in the tree
- fix get_block updated flag which was true for blocks already in cache and was leading to infinite loops on "unrelated block" errors
- apply changes to blocks in cache only after successful writes (using "virtual changes")
- do not replace cached block with an older version from disk
- recheck "unrelated blocks" (read/update collisions) until data stops changing
- track tree path correctly - do not treat split block as parent of its right half
- correctly move blocks when finding new empty place on disk
- restart updates from the beginning when one of blocks is changed by a parallel update
- fix delete using SET opcode and setting key to the empty value instead
- prevent changing the same key more than 1 time in parallel
- fix listing verification
- resume continue_updates in update_find (required because it uses continue_update itself)
- add allow_old_cached parameter to get()
2023-12-01 01:17:03 +03:00
Vitaliy Filippov 20321aaaef Implement K/V DB stress tester 2023-12-01 01:17:03 +03:00
Vitaliy Filippov 987b005356 Evict blocks based on memory limit & block usage 2023-12-01 01:17:03 +03:00
Vitaliy Filippov 41754b748b Track blocks per level 2023-12-01 01:17:03 +03:00
Vitaliy Filippov 31913256f3 Track block level 2023-12-01 01:17:03 +03:00
Vitaliy Filippov 0ee36baed7 Experimental B-Tree Vitastor embedded K/V database implementation! 2023-12-01 01:17:03 +03:00
7 changed files with 3222 additions and 5 deletions

View File

@ -181,6 +181,25 @@ target_link_libraries(vitastor-nbd
vitastor_client
)
# vitastor-kv
add_executable(vitastor-kv
kv_cli.cpp
kv_db.cpp
kv_db.h
)
target_link_libraries(vitastor-kv
vitastor_client
)
add_executable(vitastor-kv-stress
kv_stress.cpp
kv_db.cpp
kv_db.h
)
target_link_libraries(vitastor-kv-stress
vitastor_client
)
# vitastor-nfs
add_executable(vitastor-nfs
nfs_proxy.cpp

View File

@ -6,7 +6,7 @@
#include "cluster_client_impl.h"
#include "http_client.h" // json_is_true
cluster_client_t::cluster_client_t(ring_loop_t *ringloop, timerfd_manager_t *tfd, json11::Json & config)
cluster_client_t::cluster_client_t(ring_loop_t *ringloop, timerfd_manager_t *tfd, json11::Json config)
{
wb = new writeback_cache_t();
@ -532,7 +532,7 @@ void cluster_client_t::execute_internal(cluster_op_t *op)
return;
}
if (op->opcode == OSD_OP_WRITE && enable_writeback && !(op->flags & OP_FLUSH_BUFFER) &&
!op->version /* FIXME no CAS writeback */)
!op->version /* no CAS writeback */)
{
if (wb->writebacks_active >= client_max_writeback_iodepth)
{
@ -553,7 +553,7 @@ void cluster_client_t::execute_internal(cluster_op_t *op)
}
if (op->opcode == OSD_OP_WRITE && !(op->flags & OP_IMMEDIATE_COMMIT))
{
if (!(op->flags & OP_FLUSH_BUFFER))
if (!(op->flags & OP_FLUSH_BUFFER) && !op->version /* no CAS write-repeat */)
{
wb->copy_write(op, CACHE_WRITTEN);
}
@ -1152,7 +1152,7 @@ void cluster_client_t::handle_op_part(cluster_op_part_t *part)
osd_op_names[part->op.req.hdr.opcode], part->osd_num, part->op.reply.hdr.retval, expected
);
}
else
else if (log_level > 0)
{
fprintf(
stderr, "%s operation failed on OSD %lu: retval=%ld (expected %d)\n",

View File

@ -121,7 +121,7 @@ public:
json11::Json::object cli_config, file_config, etcd_global_config;
json11::Json::object config;
cluster_client_t(ring_loop_t *ringloop, timerfd_manager_t *tfd, json11::Json & config);
cluster_client_t(ring_loop_t *ringloop, timerfd_manager_t *tfd, json11::Json config);
~cluster_client_t();
void execute(cluster_op_t *op);
void execute_raw(osd_num_t osd_num, osd_op_t *op);

401
src/kv_cli.cpp Normal file
View File

@ -0,0 +1,401 @@
// Copyright (c) Vitaliy Filippov, 2019+
// License: VNPL-1.1 (see README.md for details)
//
// Vitastor shared key/value database test CLI
#define _XOPEN_SOURCE
#include <limits.h>
#include <netinet/tcp.h>
#include <sys/epoll.h>
#include <unistd.h>
#include <fcntl.h>
//#include <signal.h>
#include "epoll_manager.h"
#include "str_util.h"
#include "kv_db.h"
const char *exe_name = NULL;
class kv_cli_t
{
public:
kv_dbw_t *db = NULL;
ring_loop_t *ringloop = NULL;
epoll_manager_t *epmgr = NULL;
cluster_client_t *cli = NULL;
bool interactive = false;
int in_progress = 0;
char *cur_cmd = NULL;
int cur_cmd_size = 0, cur_cmd_alloc = 0;
bool finished = false, eof = false;
json11::Json::object cfg;
~kv_cli_t();
static json11::Json::object parse_args(int narg, const char *args[]);
void run(const json11::Json::object & cfg);
void read_cmd();
void next_cmd();
void handle_cmd(const std::string & cmd, std::function<void()> cb);
};
kv_cli_t::~kv_cli_t()
{
if (cur_cmd)
{
free(cur_cmd);
cur_cmd = NULL;
}
cur_cmd_alloc = 0;
if (db)
delete db;
if (cli)
{
cli->flush();
delete cli;
}
if (epmgr)
delete epmgr;
if (ringloop)
delete ringloop;
}
json11::Json::object kv_cli_t::parse_args(int narg, const char *args[])
{
json11::Json::object cfg;
for (int i = 1; i < narg; i++)
{
if (!strcmp(args[i], "-h") || !strcmp(args[i], "--help"))
{
printf(
"Vitastor Key/Value CLI\n"
"(c) Vitaliy Filippov, 2023+ (VNPL-1.1)\n"
"\n"
"USAGE: %s [--etcd_address ADDR] [OTHER OPTIONS]\n",
exe_name
);
exit(0);
}
else if (args[i][0] == '-' && args[i][1] == '-')
{
const char *opt = args[i]+2;
cfg[opt] = !strcmp(opt, "json") || i == narg-1 ? "1" : args[++i];
}
}
return cfg;
}
void kv_cli_t::run(const json11::Json::object & cfg)
{
// Create client
ringloop = new ring_loop_t(512);
epmgr = new epoll_manager_t(ringloop);
cli = new cluster_client_t(ringloop, epmgr->tfd, cfg);
db = new kv_dbw_t(cli);
// Load image metadata
while (!cli->is_ready())
{
ringloop->loop();
if (cli->is_ready())
break;
ringloop->wait();
}
// Run
fcntl(0, F_SETFL, fcntl(0, F_GETFL, 0) | O_NONBLOCK);
try
{
epmgr->tfd->set_fd_handler(0, false, [this](int fd, int events)
{
if (events & EPOLLIN)
{
read_cmd();
}
if (events & EPOLLRDHUP)
{
epmgr->tfd->set_fd_handler(0, false, NULL);
finished = true;
}
});
interactive = true;
printf("> ");
}
catch (std::exception & e)
{
// Can't add to epoll, STDIN is probably a file
read_cmd();
}
while (!finished)
{
ringloop->loop();
if (!finished)
ringloop->wait();
}
// Destroy the client
delete db;
db = NULL;
cli->flush();
delete cli;
delete epmgr;
delete ringloop;
cli = NULL;
epmgr = NULL;
ringloop = NULL;
}
void kv_cli_t::read_cmd()
{
if (!cur_cmd_alloc)
{
cur_cmd_alloc = 65536;
cur_cmd = (char*)malloc_or_die(cur_cmd_alloc);
}
while (cur_cmd_size < cur_cmd_alloc)
{
int r = read(0, cur_cmd+cur_cmd_size, cur_cmd_alloc-cur_cmd_size);
if (r < 0 && errno != EAGAIN)
fprintf(stderr, "Error reading from stdin: %s\n", strerror(errno));
if (r > 0)
cur_cmd_size += r;
if (r == 0)
eof = true;
if (r <= 0)
break;
}
next_cmd();
}
void kv_cli_t::next_cmd()
{
if (in_progress > 0)
{
return;
}
int pos = 0;
for (; pos < cur_cmd_size; pos++)
{
if (cur_cmd[pos] == '\n' || cur_cmd[pos] == '\r')
{
auto cmd = trim(std::string(cur_cmd, pos));
pos++;
memmove(cur_cmd, cur_cmd+pos, cur_cmd_size-pos);
cur_cmd_size -= pos;
in_progress++;
handle_cmd(cmd, [this]()
{
in_progress--;
if (interactive)
printf("> ");
next_cmd();
if (!in_progress)
read_cmd();
});
break;
}
}
if (eof && !in_progress)
{
finished = true;
}
}
void kv_cli_t::handle_cmd(const std::string & cmd, std::function<void()> cb)
{
if (cmd == "")
{
cb();
return;
}
auto pos = cmd.find_first_of(" \t");
if (pos != std::string::npos)
{
while (pos < cmd.size()-1 && (cmd[pos+1] == ' ' || cmd[pos+1] == '\t'))
pos++;
}
auto opname = strtolower(pos == std::string::npos ? cmd : cmd.substr(0, pos));
if (opname == "open")
{
uint64_t pool_id = 0;
inode_t inode_id = 0;
uint32_t kv_block_size = 0;
int scanned = sscanf(cmd.c_str() + pos+1, "%lu %lu %u", &pool_id, &inode_id, &kv_block_size);
if (scanned == 2)
{
kv_block_size = 4096;
}
if (scanned < 2 || !pool_id || !inode_id || !kv_block_size || (kv_block_size & (kv_block_size-1)) != 0)
{
fprintf(stderr, "Usage: open <pool_id> <inode_id> [block_size]. Block size must be a power of 2. Default is 4096.\n");
cb();
return;
}
cfg["kv_block_size"] = (uint64_t)kv_block_size;
db->open(INODE_WITH_POOL(pool_id, inode_id), cfg, [=](int res)
{
if (res < 0)
fprintf(stderr, "Error opening index: %s (code %d)\n", strerror(-res), res);
else
printf("Index opened. Current size: %lu bytes\n", db->get_size());
cb();
});
}
else if (opname == "config")
{
auto pos2 = cmd.find_first_of(" \t", pos+1);
if (pos2 == std::string::npos)
{
fprintf(stderr, "Usage: config <property> <value>\n");
cb();
return;
}
auto key = trim(cmd.substr(pos+1, pos2-pos-1));
auto value = parse_size(trim(cmd.substr(pos2+1)));
if (key != "kv_memory_limit" &&
key != "kv_allocate_blocks" &&
key != "kv_evict_max_misses" &&
key != "kv_evict_attempts_per_level" &&
key != "kv_evict_unused_age" &&
key != "kv_log_level")
{
fprintf(
stderr, "Allowed properties: kv_memory_limit, kv_allocate_blocks,"
" kv_evict_max_misses, kv_evict_attempts_per_level, kv_evict_unused_age, kv_log_level\n"
);
}
else
{
cfg[key] = value;
db->set_config(cfg);
}
cb();
}
else if (opname == "get" || opname == "set" || opname == "del")
{
if (opname == "get" || opname == "del")
{
if (pos == std::string::npos)
{
fprintf(stderr, "Usage: %s <key>\n", opname.c_str());
cb();
return;
}
auto key = trim(cmd.substr(pos+1));
if (opname == "get")
{
db->get(key, [this, cb](int res, const std::string & value)
{
if (res < 0)
fprintf(stderr, "Error: %s (code %d)\n", strerror(-res), res);
else
{
write(1, value.c_str(), value.size());
write(1, "\n", 1);
}
cb();
});
}
else
{
db->del(key, [this, cb](int res)
{
if (res < 0)
fprintf(stderr, "Error: %s (code %d)\n", strerror(-res), res);
else
printf("OK\n");
cb();
});
}
}
else
{
auto pos2 = cmd.find_first_of(" \t", pos+1);
if (pos2 == std::string::npos)
{
fprintf(stderr, "Usage: set <key> <value>\n");
cb();
return;
}
auto key = trim(cmd.substr(pos+1, pos2-pos-1));
auto value = trim(cmd.substr(pos2+1));
db->set(key, value, [this, cb](int res)
{
if (res < 0)
fprintf(stderr, "Error: %s (code %d)\n", strerror(-res), res);
else
printf("OK\n");
cb();
});
}
}
else if (opname == "list")
{
std::string start, end;
if (pos != std::string::npos)
{
auto pos2 = cmd.find_first_of(" \t", pos+1);
if (pos2 != std::string::npos)
{
start = trim(cmd.substr(pos+1, pos2-pos-1));
end = trim(cmd.substr(pos2+1));
}
else
{
start = trim(cmd.substr(pos+1));
}
}
void *handle = db->list_start(start);
db->list_next(handle, [=](int res, const std::string & key, const std::string & value)
{
if (res < 0)
{
if (res != -ENOENT)
{
fprintf(stderr, "Error: %s (code %d)\n", strerror(-res), res);
}
db->list_close(handle);
cb();
}
else
{
printf("%s = %s\n", key.c_str(), value.c_str());
db->list_next(handle, NULL);
}
});
}
else if (opname == "close")
{
db->close([=]()
{
printf("Index closed\n");
cb();
});
}
else if (opname == "quit" || opname == "q")
{
::close(0);
finished = true;
}
else
{
fprintf(
stderr, "Unknown operation: %s. Supported operations:\n"
"open <pool_id> <inode_id> [block_size]\n"
"config <property> <value>\n"
"get <key>\nset <key> <value>\ndel <key>\nlist [<start> [end]]\n"
"close\nquit\n", opname.c_str()
);
cb();
}
}
int main(int narg, const char *args[])
{
setvbuf(stdout, NULL, _IONBF, 0);
setvbuf(stderr, NULL, _IONBF, 0);
exe_name = args[0];
kv_cli_t *p = new kv_cli_t();
p->run(kv_cli_t::parse_args(narg, args));
delete p;
return 0;
}

2064
src/kv_db.cpp Normal file

File diff suppressed because it is too large Load Diff

36
src/kv_db.h Normal file
View File

@ -0,0 +1,36 @@
// Copyright (c) Vitaliy Filippov, 2019+
// License: VNPL-1.1 (see README.md for details)
//
// Vitastor shared key/value database
// Parallel optimistic B-Tree O:-)
#pragma once
#include "cluster_client.h"
struct kv_db_t;
struct kv_dbw_t
{
kv_dbw_t(cluster_client_t *cli);
~kv_dbw_t();
void open(inode_t inode_id, json11::Json cfg, std::function<void(int)> cb);
void set_config(json11::Json cfg);
void close(std::function<void()> cb);
uint64_t get_size();
void get(const std::string & key, std::function<void(int res, const std::string & value)> cb,
bool allow_old_cached = false);
void set(const std::string & key, const std::string & value, std::function<void(int res)> cb,
std::function<bool(int res, const std::string & value)> cas_compare = NULL);
void del(const std::string & key, std::function<void(int res)> cb,
std::function<bool(int res, const std::string & value)> cas_compare = NULL);
void* list_start(const std::string & start);
void list_next(void *handle, std::function<void(int res, const std::string & key, const std::string & value)> cb);
void list_close(void *handle);
kv_db_t *db;
};

697
src/kv_stress.cpp Normal file
View File

@ -0,0 +1,697 @@
// Copyright (c) Vitaliy Filippov, 2019+
// License: VNPL-1.1 (see README.md for details)
//
// Vitastor shared key/value database stress tester / benchmark
#define _XOPEN_SOURCE
#include <limits.h>
#include <netinet/tcp.h>
#include <sys/epoll.h>
#include <unistd.h>
#include <fcntl.h>
//#include <signal.h>
#include "epoll_manager.h"
#include "str_util.h"
#include "kv_db.h"
const char *exe_name = NULL;
struct kv_test_listing_t
{
uint64_t count = 0, done = 0;
void *handle = NULL;
std::string next_after;
std::set<std::string> inflights;
timespec tv_begin;
bool error = false;
};
struct kv_test_lat_t
{
const char *name = NULL;
uint64_t usec = 0, count = 0;
};
struct kv_test_stat_t
{
kv_test_lat_t get, add, update, del, list;
uint64_t list_keys = 0;
};
class kv_test_t
{
public:
// Config
json11::Json::object kv_cfg;
std::string key_prefix, key_suffix;
uint64_t inode_id = 0;
uint64_t op_count = 1000000;
uint64_t runtime_sec = 0;
uint64_t parallelism = 4;
uint64_t reopen_prob = 1;
uint64_t get_prob = 30000;
uint64_t add_prob = 20000;
uint64_t update_prob = 20000;
uint64_t del_prob = 5000;
uint64_t list_prob = 300;
uint64_t min_key_len = 10;
uint64_t max_key_len = 70;
uint64_t min_value_len = 50;
uint64_t max_value_len = 300;
uint64_t min_list_count = 10;
uint64_t max_list_count = 1000;
uint64_t print_stats_interval = 1;
bool json_output = false;
uint64_t log_level = 1;
bool trace = false;
bool stop_on_error = false;
// FIXME: Multiple clients
kv_test_stat_t stat, prev_stat;
timespec prev_stat_time, start_stat_time;
// State
kv_dbw_t *db = NULL;
ring_loop_t *ringloop = NULL;
epoll_manager_t *epmgr = NULL;
cluster_client_t *cli = NULL;
ring_consumer_t consumer;
bool finished = false;
uint64_t total_prob = 0;
uint64_t ops_sent = 0, ops_done = 0;
int stat_timer_id = -1;
int in_progress = 0;
bool reopening = false;
std::set<kv_test_listing_t*> listings;
std::set<std::string> changing_keys;
std::map<std::string, std::string> values;
~kv_test_t();
static json11::Json::object parse_args(int narg, const char *args[]);
void parse_config(json11::Json cfg);
void run(json11::Json cfg);
void loop();
void print_stats(kv_test_stat_t & prev_stat, timespec & prev_stat_time);
void print_total_stats();
void start_change(const std::string & key);
void stop_change(const std::string & key);
void add_stat(kv_test_lat_t & stat, timespec tv_begin);
};
kv_test_t::~kv_test_t()
{
if (db)
delete db;
if (cli)
{
cli->flush();
delete cli;
}
if (epmgr)
delete epmgr;
if (ringloop)
delete ringloop;
}
json11::Json::object kv_test_t::parse_args(int narg, const char *args[])
{
json11::Json::object cfg;
for (int i = 1; i < narg; i++)
{
if (!strcmp(args[i], "-h") || !strcmp(args[i], "--help"))
{
printf(
"Vitastor Key/Value DB stress tester / benchmark\n"
"(c) Vitaliy Filippov, 2023+ (VNPL-1.1)\n"
"\n"
"USAGE: %s --pool_id POOL_ID --inode_id INODE_ID [OPTIONS]\n"
" --op_count 1000000\n"
" Total operations to run during test. 0 means unlimited\n"
" --key_prefix \"\"\n"
" Prefix for all keys read or written (to avoid collisions)\n"
" --key_suffix \"\"\n"
" Suffix for all keys read or written (to avoid collisions, but scan all DB)\n"
" --runtime 0\n"
" Run for this number of seconds. 0 means unlimited\n"
" --parallelism 4\n"
" Run this number of operations in parallel\n"
" --get_prob 30000\n"
" Fraction of key retrieve operations\n"
" --add_prob 20000\n"
" Fraction of key addition operations\n"
" --update_prob 20000\n"
" Fraction of key update operations\n"
" --del_prob 30000\n"
" Fraction of key delete operations\n"
" --list_prob 300\n"
" Fraction of listing operations\n"
" --min_key_len 10\n"
" Minimum key size in bytes\n"
" --max_key_len 70\n"
" Maximum key size in bytes\n"
" --min_value_len 50\n"
" Minimum value size in bytes\n"
" --max_value_len 300\n"
" Maximum value size in bytes\n"
" --min_list_count 10\n"
" Minimum number of keys read in listing (0 = all keys)\n"
" --max_list_count 1000\n"
" Maximum number of keys read in listing\n"
" --print_stats 1\n"
" Print operation statistics every this number of seconds\n"
" --json\n"
" JSON output\n"
" --stop_on_error 0\n"
" Stop on first execution error, mismatch, lost key or extra key during listing\n"
" --kv_memory_limit 128M\n"
" Maximum memory to use for vitastor-kv index cache\n"
" --kv_allocate_blocks 4\n"
" Number of PG blocks used for new tree block allocation in parallel\n"
" --kv_evict_max_misses 10\n"
" Eviction algorithm parameter: retry eviction from another random spot\n"
" if this number of keys is used currently or was used recently\n"
" --kv_evict_attempts_per_level 3\n"
" Retry eviction at most this number of times per tree level, starting\n"
" with bottom-most levels\n"
" --kv_evict_unused_age 1000\n"
" Evict only keys unused during this number of last operations\n"
" --kv_log_level 1\n"
" Log level. 0 = errors, 1 = warnings, 10 = trace operations\n",
exe_name
);
exit(0);
}
else if (args[i][0] == '-' && args[i][1] == '-')
{
const char *opt = args[i]+2;
cfg[opt] = !strcmp(opt, "json") || i == narg-1 ? "1" : args[++i];
}
}
return cfg;
}
void kv_test_t::parse_config(json11::Json cfg)
{
inode_id = INODE_WITH_POOL(cfg["pool_id"].uint64_value(), cfg["inode_id"].uint64_value());
if (cfg["op_count"].uint64_value() > 0)
op_count = cfg["op_count"].uint64_value();
key_prefix = cfg["key_prefix"].string_value();
key_suffix = cfg["key_suffix"].string_value();
if (cfg["runtime"].uint64_value() > 0)
runtime_sec = cfg["runtime"].uint64_value();
if (cfg["parallelism"].uint64_value() > 0)
parallelism = cfg["parallelism"].uint64_value();
if (!cfg["reopen_prob"].is_null())
reopen_prob = cfg["reopen_prob"].uint64_value();
if (!cfg["get_prob"].is_null())
get_prob = cfg["get_prob"].uint64_value();
if (!cfg["add_prob"].is_null())
add_prob = cfg["add_prob"].uint64_value();
if (!cfg["update_prob"].is_null())
update_prob = cfg["update_prob"].uint64_value();
if (!cfg["del_prob"].is_null())
del_prob = cfg["del_prob"].uint64_value();
if (!cfg["list_prob"].is_null())
list_prob = cfg["list_prob"].uint64_value();
if (!cfg["min_key_len"].is_null())
min_key_len = cfg["min_key_len"].uint64_value();
if (cfg["max_key_len"].uint64_value() > 0)
max_key_len = cfg["max_key_len"].uint64_value();
if (!cfg["min_value_len"].is_null())
min_value_len = cfg["min_value_len"].uint64_value();
if (cfg["max_value_len"].uint64_value() > 0)
max_value_len = cfg["max_value_len"].uint64_value();
if (!cfg["min_list_count"].is_null())
min_list_count = cfg["min_list_count"].uint64_value();
if (!cfg["max_list_count"].is_null())
max_list_count = cfg["max_list_count"].uint64_value();
if (!cfg["print_stats"].is_null())
print_stats_interval = cfg["print_stats"].uint64_value();
if (!cfg["json"].is_null())
json_output = true;
if (!cfg["stop_on_error"].is_null())
stop_on_error = cfg["stop_on_error"].bool_value();
if (!cfg["kv_memory_limit"].is_null())
kv_cfg["kv_memory_limit"] = cfg["kv_memory_limit"];
if (!cfg["kv_allocate_blocks"].is_null())
kv_cfg["kv_allocate_blocks"] = cfg["kv_allocate_blocks"];
if (!cfg["kv_evict_max_misses"].is_null())
kv_cfg["kv_evict_max_misses"] = cfg["kv_evict_max_misses"];
if (!cfg["kv_evict_attempts_per_level"].is_null())
kv_cfg["kv_evict_attempts_per_level"] = cfg["kv_evict_attempts_per_level"];
if (!cfg["kv_evict_unused_age"].is_null())
kv_cfg["kv_evict_unused_age"] = cfg["kv_evict_unused_age"];
if (!cfg["kv_log_level"].is_null())
{
log_level = cfg["kv_log_level"].uint64_value();
trace = log_level >= 10;
kv_cfg["kv_log_level"] = cfg["kv_log_level"];
}
total_prob = reopen_prob+get_prob+add_prob+update_prob+del_prob+list_prob;
stat.get.name = "get";
stat.add.name = "add";
stat.update.name = "update";
stat.del.name = "del";
stat.list.name = "list";
}
void kv_test_t::run(json11::Json cfg)
{
srand48(time(NULL));
parse_config(cfg);
// Create client
ringloop = new ring_loop_t(512);
epmgr = new epoll_manager_t(ringloop);
cli = new cluster_client_t(ringloop, epmgr->tfd, cfg);
db = new kv_dbw_t(cli);
// Load image metadata
while (!cli->is_ready())
{
ringloop->loop();
if (cli->is_ready())
break;
ringloop->wait();
}
// Run
reopening = true;
db->open(inode_id, kv_cfg, [this](int res)
{
reopening = false;
if (res < 0)
{
fprintf(stderr, "ERROR: Open index: %d (%s)\n", res, strerror(-res));
exit(1);
}
if (trace)
printf("Index opened\n");
ringloop->wakeup();
});
consumer.loop = [this]() { loop(); };
ringloop->register_consumer(&consumer);
if (print_stats_interval)
stat_timer_id = epmgr->tfd->set_timer(print_stats_interval*1000, true, [this](int) { print_stats(prev_stat, prev_stat_time); });
clock_gettime(CLOCK_REALTIME, &start_stat_time);
prev_stat_time = start_stat_time;
while (!finished)
{
ringloop->loop();
if (!finished)
ringloop->wait();
}
if (stat_timer_id >= 0)
epmgr->tfd->clear_timer(stat_timer_id);
ringloop->unregister_consumer(&consumer);
// Print total stats
print_total_stats();
// Destroy the client
delete db;
db = NULL;
cli->flush();
delete cli;
delete epmgr;
delete ringloop;
cli = NULL;
epmgr = NULL;
ringloop = NULL;
}
static const char *base64_chars = "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789@+/";
std::string random_str(int len)
{
std::string str;
str.resize(len);
for (int i = 0; i < len; i++)
{
str[i] = base64_chars[lrand48() % 64];
}
return str;
}
void kv_test_t::loop()
{
if (reopening)
{
return;
}
if (ops_done >= op_count)
{
finished = true;
}
while (!finished && ops_sent < op_count && in_progress < parallelism)
{
uint64_t dice = (lrand48() % total_prob);
if (dice < reopen_prob)
{
reopening = true;
db->close([this]()
{
if (trace)
printf("Index closed\n");
db->open(inode_id, kv_cfg, [this](int res)
{
reopening = false;
if (res < 0)
{
fprintf(stderr, "ERROR: Reopen index: %d (%s)\n", res, strerror(-res));
finished = true;
return;
}
if (trace)
printf("Index reopened\n");
ringloop->wakeup();
});
});
return;
}
else if (dice < reopen_prob+get_prob)
{
// get existing
auto key = random_str(max_key_len);
auto k_it = values.lower_bound(key);
if (k_it == values.end())
continue;
key = k_it->first;
if (changing_keys.find(key) != changing_keys.end())
continue;
in_progress++;
ops_sent++;
if (trace)
printf("get %s\n", key.c_str());
timespec tv_begin;
clock_gettime(CLOCK_REALTIME, &tv_begin);
db->get(key, [this, key, tv_begin](int res, const std::string & value)
{
add_stat(stat.get, tv_begin);
ops_done++;
in_progress--;
auto it = values.find(key);
if (res != (it == values.end() ? -ENOENT : 0))
{
fprintf(stderr, "ERROR: get %s: %d (%s)\n", key.c_str(), res, strerror(-res));
if (stop_on_error)
exit(1);
}
else if (it != values.end() && value != it->second)
{
fprintf(stderr, "ERROR: get %s: mismatch: %s vs %s\n", key.c_str(), value.c_str(), it->second.c_str());
if (stop_on_error)
exit(1);
}
ringloop->wakeup();
});
}
else if (dice < reopen_prob+get_prob+add_prob+update_prob)
{
bool is_add = false;
std::string key;
if (dice < reopen_prob+get_prob+add_prob)
{
// add
is_add = true;
uint64_t key_len = min_key_len + (max_key_len > min_key_len ? lrand48() % (max_key_len-min_key_len) : 0);
key = key_prefix + random_str(key_len) + key_suffix;
}
else
{
// update
key = random_str(max_key_len);
auto k_it = values.lower_bound(key);
if (k_it == values.end())
continue;
key = k_it->first;
}
if (changing_keys.find(key) != changing_keys.end())
continue;
uint64_t value_len = min_value_len + (max_value_len > min_value_len ? lrand48() % (max_value_len-min_value_len) : 0);
auto value = random_str(value_len);
start_change(key);
ops_sent++;
in_progress++;
if (trace)
printf("set %s = %s\n", key.c_str(), value.c_str());
timespec tv_begin;
clock_gettime(CLOCK_REALTIME, &tv_begin);
db->set(key, value, [this, key, value, tv_begin, is_add](int res)
{
add_stat(is_add ? stat.add : stat.update, tv_begin);
stop_change(key);
ops_done++;
in_progress--;
if (res != 0)
{
fprintf(stderr, "ERROR: set %s = %s: %d (%s)\n", key.c_str(), value.c_str(), res, strerror(-res));
if (stop_on_error)
exit(1);
}
else
{
values[key] = value;
}
ringloop->wakeup();
}, NULL);
}
else if (dice < reopen_prob+get_prob+add_prob+update_prob+del_prob)
{
// delete
auto key = random_str(max_key_len);
auto k_it = values.lower_bound(key);
if (k_it == values.end())
continue;
key = k_it->first;
if (changing_keys.find(key) != changing_keys.end())
continue;
start_change(key);
ops_sent++;
in_progress++;
if (trace)
printf("del %s\n", key.c_str());
timespec tv_begin;
clock_gettime(CLOCK_REALTIME, &tv_begin);
db->del(key, [this, key, tv_begin](int res)
{
add_stat(stat.del, tv_begin);
stop_change(key);
ops_done++;
in_progress--;
if (res != 0)
{
fprintf(stderr, "ERROR: del %s: %d (%s)\n", key.c_str(), res, strerror(-res));
if (stop_on_error)
exit(1);
}
else
{
values.erase(key);
}
ringloop->wakeup();
}, NULL);
}
else if (dice < reopen_prob+get_prob+add_prob+update_prob+del_prob+list_prob)
{
// list
ops_sent++;
in_progress++;
auto key = random_str(max_key_len);
auto lst = new kv_test_listing_t;
auto k_it = values.lower_bound(key);
lst->count = min_list_count + (max_list_count > min_list_count ? lrand48() % (max_list_count-min_list_count) : 0);
lst->handle = db->list_start(k_it == values.begin() ? key_prefix : key);
lst->next_after = k_it == values.begin() ? key_prefix : key;
lst->inflights = changing_keys;
listings.insert(lst);
if (trace)
printf("list from %s\n", key.c_str());
clock_gettime(CLOCK_REALTIME, &lst->tv_begin);
db->list_next(lst->handle, [this, lst](int res, const std::string & key, const std::string & value)
{
if (log_level >= 11)
printf("list: %s = %s\n", key.c_str(), value.c_str());
if (res >= 0 && key_prefix.size() && (key.size() < key_prefix.size() ||
key.substr(0, key_prefix.size()) != key_prefix))
{
// stop at this key
res = -ENOENT;
}
if (res < 0 || (lst->count > 0 && lst->done >= lst->count))
{
add_stat(stat.list, lst->tv_begin);
if (res == 0)
{
// ok (done >= count)
}
else if (res != -ENOENT)
{
fprintf(stderr, "ERROR: list: %d (%s)\n", res, strerror(-res));
lst->error = true;
}
else
{
auto k_it = lst->next_after == "" ? values.begin() : values.upper_bound(lst->next_after);
while (k_it != values.end())
{
while (k_it != values.end() && lst->inflights.find(k_it->first) != lst->inflights.end())
k_it++;
if (k_it != values.end())
{
fprintf(stderr, "ERROR: list: missing key %s\n", (k_it++)->first.c_str());
lst->error = true;
}
}
}
if (lst->error && stop_on_error)
exit(1);
ops_done++;
in_progress--;
db->list_close(lst->handle);
delete lst;
listings.erase(lst);
ringloop->wakeup();
}
else
{
stat.list_keys++;
// Do not check modified keys in listing
// Listing may return their old or new state
if ((!key_suffix.size() || key.size() >= key_suffix.size() &&
key.substr(key.size()-key_suffix.size()) == key_suffix) &&
lst->inflights.find(key) == lst->inflights.end())
{
lst->done++;
auto k_it = lst->next_after == "" ? values.begin() : values.upper_bound(lst->next_after);
while (true)
{
while (k_it != values.end() && lst->inflights.find(k_it->first) != lst->inflights.end())
{
k_it++;
}
if (k_it == values.end() || k_it->first > key)
{
fprintf(stderr, "ERROR: list: extra key %s\n", key.c_str());
lst->error = true;
break;
}
else if (k_it->first < key)
{
fprintf(stderr, "ERROR: list: missing key %s\n", k_it->first.c_str());
lst->error = true;
lst->next_after = k_it->first;
k_it++;
}
else
{
if (k_it->second != value)
{
fprintf(stderr, "ERROR: list: mismatch: %s = %s but should be %s\n",
key.c_str(), value.c_str(), k_it->second.c_str());
lst->error = true;
}
lst->next_after = k_it->first;
break;
}
}
}
db->list_next(lst->handle, NULL);
}
});
}
}
}
void kv_test_t::add_stat(kv_test_lat_t & stat, timespec tv_begin)
{
timespec tv_end;
clock_gettime(CLOCK_REALTIME, &tv_end);
int64_t usec = (tv_end.tv_sec - tv_begin.tv_sec)*1000000 +
(tv_end.tv_nsec - tv_begin.tv_nsec)/1000;
if (usec > 0)
{
stat.usec += usec;
stat.count++;
}
}
void kv_test_t::print_stats(kv_test_stat_t & prev_stat, timespec & prev_stat_time)
{
timespec cur_stat_time;
clock_gettime(CLOCK_REALTIME, &cur_stat_time);
int64_t usec = (cur_stat_time.tv_sec - prev_stat_time.tv_sec)*1000000 +
(cur_stat_time.tv_nsec - prev_stat_time.tv_nsec)/1000;
if (usec > 0)
{
kv_test_lat_t *lats[] = { &stat.get, &stat.add, &stat.update, &stat.del, &stat.list };
kv_test_lat_t *prev[] = { &prev_stat.get, &prev_stat.add, &prev_stat.update, &prev_stat.del, &prev_stat.list };
if (!json_output)
{
char buf[128] = { 0 };
for (int i = 0; i < sizeof(lats)/sizeof(lats[0]); i++)
{
snprintf(buf, sizeof(buf)-1, "%.1f %s/s (%lu us)", (lats[i]->count-prev[i]->count)*1000000.0/usec,
lats[i]->name, (lats[i]->usec-prev[i]->usec)/(lats[i]->count-prev[i]->count > 0 ? lats[i]->count-prev[i]->count : 1));
int k;
for (k = strlen(buf); k < strlen(lats[i]->name)+21; k++)
buf[k] = ' ';
buf[k] = 0;
printf("%s", buf);
}
printf("\n");
}
else
{
int64_t runtime = (cur_stat_time.tv_sec - start_stat_time.tv_sec)*1000000 +
(cur_stat_time.tv_nsec - start_stat_time.tv_nsec)/1000;
printf("{\"runtime\":%.1f", (double)runtime/1000000.0);
for (int i = 0; i < sizeof(lats)/sizeof(lats[0]); i++)
{
if (lats[i]->count > prev[i]->count)
{
printf(
",\"%s\":{\"avg\":{\"iops\":%.1f,\"usec\":%lu},\"total\":{\"count\":%lu,\"usec\":%lu}}",
lats[i]->name, (lats[i]->count-prev[i]->count)*1000000.0/usec,
(lats[i]->usec-prev[i]->usec)/(lats[i]->count-prev[i]->count),
lats[i]->count, lats[i]->usec
);
}
}
printf("}\n");
}
}
prev_stat = stat;
prev_stat_time = cur_stat_time;
}
void kv_test_t::print_total_stats()
{
if (!json_output)
printf("Total:\n");
kv_test_stat_t start_stats;
timespec start_stat_time = this->start_stat_time;
print_stats(start_stats, start_stat_time);
}
void kv_test_t::start_change(const std::string & key)
{
changing_keys.insert(key);
for (auto lst: listings)
{
lst->inflights.insert(key);
}
}
void kv_test_t::stop_change(const std::string & key)
{
changing_keys.erase(key);
}
int main(int narg, const char *args[])
{
setvbuf(stdout, NULL, _IONBF, 0);
setvbuf(stderr, NULL, _IONBF, 0);
exe_name = args[0];
kv_test_t *p = new kv_test_t();
p->run(kv_test_t::parse_args(narg, args));
delete p;
return 0;
}