Repository navigation
perf: 3x smaller shared stats copies, 2-field copies on most messages - #1993
Merged
Merged
Conversation
Codecov Report✅ All modified and coverable lines are covered by tests. 📢 Thoughts on this report? Let us know! |
jkaczman
force-pushed
the
jk-server-stats-optim
branch
from
November 8, 2026 22:48
faa04c9 to
27a26e6
Compare
Contributor
|
Metal bench (arm64) failed. 2 of 475 measures outside bounds.
Counters per txn
Microbench: 460 of 462 measures within bounds.
Microbench latency: 4 benches changed more than 3%
Hot code per txn
pooler:
sharding-w2:
Logslb, microbench, pooler, sharding-w2, sharding-w8, verdict.json |
Contributor
|
Metal bench (x86) passed. All 475 measures within bounds.
Counters per txn
Microbench: all 462 measures within bounds.
Microbench latency: 11 benches changed more than 3%
Hot code per txn
pooler:
sharding-w2:
Logslb, microbench, pooler, sharding-w2, sharding-w8, verdict.json |
Contributor
|
Metal bench PGO (arm64). 6 rounds of eaab055 vs d653c73.
Compared d653c73 with eaab055 on c8g.metal-24xl (on-demand). |
Contributor
|
Metal bench PGO (x86). 6 rounds of eaab055 vs d653c73.
|
jkaczman
force-pushed
the
jk-server-stats-optim
branch
from
November 9, 2026 16:33
27a26e6 to
d653c73
Compare
levkk
approved these changes
Nov 9, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Background:
SHOW SERVERS) and metrics use as the source of their dataSHOW SERVERSshows the current last message codes (last_sentandlast_received), e.g., to debug stuck connections.SHOW SERVERSwas added (way back when!), the shared and per-connection stats BOTH used a singular type (ServerStats) to store their stats, so every single one of those shared copies unnecessarily included some fields only relevant to the per-connection store (e.g., thelast_checkoutcounters and the timers), meaning that there were 448 extra bytes being copied.The stats are copied with
memcpy, which uses 8 instructions for every 64 bytes copied. If you take 448 bytes, divide by 64, multiply by 8, that's 56 instructions for every single copy performed. The stats are copied on ALL messages to and from Postgres (e.g., once per row), so these copies add up fast!The changes below cut our instructions per request by:
Now,
SharedServerStats), which only includes fields actually used (the 216 bytes)SharedServerStatsother than those two (e.g.,query()adds 1 toqueries), it has to go throughshared_mut(), which sets adirtyflag. If the flag is set, the next message copies all 216 bytes. As it's more rare, this further improves performance for most cases.Also, we previously counted 18 counters (e.g.,
parseandrows_inserted) twice, once since the last checkout (last_checkout, whichSHOW STATSadds up at check-in) and once over the life of the connection (total). However, only tests were using the counts intotal, so I fixed the redundancy.