distsql: run last processor from main goroutine #27899
Conversation
|
This change is |
There was a problem hiding this comment.
Reviewable status:
complete! 0 of 0 LGTMs obtained
pkg/sql/distsqlrun/flow.go, line 469 at r1 (raw file):
// setup error is pushed to the syncFlowConsumer. In this case, a subsequent // call to f.Wait() will not block. func (f *Flow) Start(ctx context.Context, block bool, doneFn func()) error {
Rather than add a boolean parameter, I'd recommend adding a new method called StartSync and renaming this one to StartAsync, or similar. That will make it much more clear in each of the call sites what's actually going on.
There was a problem hiding this comment.
Reviewable status:
pkg/sql/distsqlrun/flow.go, line 469 at r1 (raw file):
Previously, jordanlewis (Jordan Lewis) wrote…
Rather than add a boolean parameter, I'd recommend adding a new method called StartSync and renaming this one to StartAsync, or similar. That will make it much more clear in each of the call sites what's actually going on.
Good call. Done.
asubiotto
force-pushed
the
gperf
branch
3 times, most recently
from
9605bf2
to
8999200
Compare
Closes #27734 This change improves distsql throughput for short-running queries by removing some synchronization overhead. On a kv --read-percent=100 workload, this change resulted in a 25% throughput improvement. Release note (performance improvement): Improve fixed cost of running distributed sql queries.
|
bors r+ |
27899: distsql: run last processor from main goroutine r=asubiotto a=asubiotto Closes #27734 This change improves distsql throughput for short-running queries by removing some synchronization overhead. On a kv --read-percent=100 workload, this change resulted in a 25% throughput improvement (however note that these numbers vary a fair amount). Release note (performance improvement): Improve fixed cost of running distributed sql queries. ``` _elapsed___errors_____ops(total)___ops/sec(cum)__avg(ms)__p50(ms)__p95(ms)__p99(ms)_pMax(ms)__result 60.0s 0 690447 11507.2 1.4 1.1 3.3 5.8 130.0 ``` ``` _elapsed___errors_____ops(total)___ops/sec(cum)__avg(ms)__p50(ms)__p95(ms)__p99(ms)_pMax(ms)__result 60.0s 0 865636 14427.7 1.1 0.9 2.6 4.2 39.8 ``` Co-authored-by: Alfonso Subiotto Marqués <alfonso@cockroachlabs.com>
Build succeeded |
Closes #27734
This change improves distsql throughput for short-running queries by
removing some synchronization overhead. On a kv --read-percent=100
workload, this change resulted in a 25% throughput improvement (however note that these numbers vary a fair amount).
Release note (performance improvement): Improve fixed cost of running
distributed sql queries.