mirrors/scylladb
1
0
Fork 0
mirror of https://github.com/scylladb/scylladb.git synced 2026-05-12 19:02:12 +00:00
Code Issues Packages Projects Releases Wiki Activity

db: view: run local materialized view mutations on a separate smp service group

Browse Source 
When base write triggers mv write and it needs to be send to another
shard it used the same service group and we could end up with a
deadlock.

This fix affects also alternator's secondary indexes.

Testing was done using (yet) not committed framework for easy alternator
performance testing: https://github.com/scylladb/scylladb/pull/13121.
I've changed hardcoded max_nonlocal_requests config in scylla from 5000 to 500 and
then ran:

./build/release/scylla perf-alternator-workloads --workdir /tmp/scylla-workdir/ --smp 2 \
--developer-mode 1 --alternator-port 8000 --alternator-write-isolation forbid --workload write_gsi \
--duration 60 --ring-delay-ms 0 --skip-wait-for-gossip-to-settle 0 --continue-after-error true --concurrency 2000

Without the patch when scylla is overloaded (i.e. number of scheduled futures being close to max_nonlocal_requests) after couple seconds
scylla hangs, cpu usage drops to zero, no progress is made. We can confirm we're hitting this issue by seeing under gdb:

p seastar::get_smp_service_groups_semaphore(2,0)._count
$1 = 0

With the patch I wasn't able to observe the problem, even with 2x
concurrency. I was able to make the process hang with 10x concurrency
but I think it's hitting different limit as there wasn't any depleted
smp service group semaphore and it was happening also on non mv loads.

Fixes https://github.com/scylladb/scylladb/issues/15844

Closes scylladb/scylladb#15845

(cherry picked from commit 020a9c931b)
This commit is contained in:
Marcin Maliszkiewicz
2023-10-25 19:46:52 +02:00
committed by Avi Kivity
parent f886581bee
commit abf62e5b7f
4 changed files with 10 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
db/view/view.cc
View File

@@ -1331,7 +1331,7 @@ future<> mutate_MV(
auto mut_ptr = remote_endpoints.empty() ? std::make_unique<frozen_mutation>(std::move(mut.fm)) : std::make_unique<frozen_mutation>(mut.fm);
tracing::trace(tr_state, "Locally applying view update for {}.{}; base token = {}; view token = {}",
mut.s->ks_name(), mut.s->cf_name(), base_token, view_token);
local_view_update = service::get_local_storage_proxy().mutate_locally(mut.s, *mut_ptr, tr_state, db::commitlog::force_sync::no).then_wrapped(
local_view_update = service::get_local_storage_proxy().mutate_mv_locally(mut.s, *mut_ptr, tr_state, db::commitlog::force_sync::no).then_wrapped(
[s = mut.s, &stats, &cf_stats, tr_state, base_token, view_token, my_address, mut_ptr = std::move(mut_ptr),
units = sem_units.split(sem_units.count())] (future<>&& f) {
--stats.writes;

1
main.cc
View File

@@ -1037,6 +1037,7 @@ To start the scylla server proper, simply invoke as: scylla server (or just scyl
storage_proxy_smp_service_group_config.max_nonlocal_requests = 5000;
spcfg.read_smp_service_group = create_smp_service_group(storage_proxy_smp_service_group_config).get0();
spcfg.write_smp_service_group = create_smp_service_group(storage_proxy_smp_service_group_config).get0();
spcfg.write_mv_smp_service_group = create_smp_service_group(storage_proxy_smp_service_group_config).get0();
spcfg.hints_write_smp_service_group = create_smp_service_group(storage_proxy_smp_service_group_config).get0();
spcfg.write_ack_smp_service_group = create_smp_service_group(storage_proxy_smp_service_group_config).get0();
static db::view::node_update_backlog node_backlog(smp::count, 10ms);

1
service/storage_proxy.cc
View File

@@ -1876,6 +1876,7 @@ storage_proxy::storage_proxy(distributed<replica::database>& db, gms::gossiper&
, _erm_factory(erm_factory)
, _read_smp_service_group(cfg.read_smp_service_group)
, _write_smp_service_group(cfg.write_smp_service_group)
, _write_mv_smp_service_group(cfg.write_mv_smp_service_group)
, _hints_write_smp_service_group(cfg.hints_write_smp_service_group)
, _write_ack_smp_service_group(cfg.write_ack_smp_service_group)
, _next_response_id(std::chrono::system_clock::now().time_since_epoch()/1ms)

7
service/storage_proxy.hh
View File

@@ -139,6 +139,7 @@ public:
size_t available_memory;
smp_service_group read_smp_service_group = default_smp_service_group();
smp_service_group write_smp_service_group = default_smp_service_group();
smp_service_group write_mv_smp_service_group = default_smp_service_group();
smp_service_group hints_write_smp_service_group = default_smp_service_group();
// Write acknowledgments might not be received on the correct shard, and
// they need a separate smp_service_group to prevent an ABBA deadlock
@@ -236,6 +237,7 @@ private:
shared_ptr<migration_manager> _mm;
smp_service_group _read_smp_service_group;
smp_service_group _write_smp_service_group;
smp_service_group _write_mv_smp_service_group;
smp_service_group _hints_write_smp_service_group;
smp_service_group _write_ack_smp_service_group;
response_id_type _next_response_id;
@@ -524,6 +526,11 @@ public:
future<> mutate_locally(const schema_ptr& s, const frozen_mutation& m, tracing::trace_state_ptr tr_state, db::commitlog::force_sync sync, clock_type::time_point timeout = clock_type::time_point::max(), db::per_partition_rate_limit::info rate_limit_info = std::monostate()) {
return mutate_locally(s, m, tr_state, sync, timeout, _write_smp_service_group, rate_limit_info);
}
// Applies materialized view mutation on this node.
// Resolves with timed_out_error when timeout is reached.
future<> mutate_mv_locally(const schema_ptr& s, const frozen_mutation& m, tracing::trace_state_ptr tr_state, db::commitlog::force_sync sync, clock_type::time_point timeout = clock_type::time_point::max(), db::per_partition_rate_limit::info rate_limit_info = std::monostate()) {
return mutate_locally(s, m, tr_state, sync, timeout, _write_mv_smp_service_group, rate_limit_info);
}
// Applies mutations on this node.
// Resolves with timed_out_error when timeout is reached.
future<> mutate_locally(std::vector<mutation> mutation, tracing::trace_state_ptr tr_state, clock_type::time_point timeout = clock_type::time_point::max(), db::per_partition_rate_limit::info rate_limit_info = std::monostate());