• Home
  • Features
  • Pricing
  • Docs
  • Announcements
  • Sign In

supabase / etl / 44213223634
79%

Build:
DEFAULT BRANCH: main
Ran 18 Nov 2026 01:14AM UTC
Jobs 1
Files 270
Run time 1min
Badge
Embed ▾
README BADGES
x

If you need to use a raster PNG badge, change the '.svg' to '.png' in the link

Markdown

Textile

RDoc

HTML

Rst

18 Nov 2026 01:06AM UTC coverage: 78.504% (-0.5%) from 78.994%
44213223634

push

github

web-flow
perf: trim lock overhead and profile tokio lock waits (#1263)

* perf: trim lock overhead and unify poisoning

PIPE-1358 asked for low-risk lock wins and one poisoning policy.

- Cache the durability age gauge at construction. Each dispatch and
  confirmation previously built three label strings and looked up the
  series under the state lock; a release-mode loop measured ~135 ns
  per update before and ~13 ns after (lock and age included). The
  gauge is still set under the lock, so late updates cannot overwrite
  a newer age or the shutdown sentinel.
- Recover poisoned sync locks with PoisonError::into_inner in the lag
  monitor, durability monitor, slot-creation monitor, and DuckLake
  interrupt registry. Their critical sections always leave valid data,
  and the old unwrap in Drop impls could escalate a panic to an abort.
- Use std mutexes for MemoryStore and the CTID copy work queue; neither
  holds its guard across an await.
- Document the poisoning policy in AGENTS.md. DuckDB connection
  mutexes keep mapping poisoning to InvalidState.

* bench: add tokio lock contention profile

Adds `nix run .#lock-contention`, a one-command profile of how long
tasks wait for tokio Mutex, RwLock, and Semaphore acquires. Thread
profilers cannot see these waits because a waiting task does not
block a thread.

- lock_contention benchmark: copies tables with PostgresStore while
  writers keep CDC running, then reports waits per lock (creation
  site and operation) with p50/p99/max and the waiting call sites.
- A tracing layer reads tokio's tokio_unstable instrumentation:
  an acquire's poll span lives from first poll to guard handoff,
  and more than one poll means the task waited.
- scripts/lock-contention.sh starts a throwaway Postgres with logical
  replication, builds with --cfg tokio_unstable, runs, and stops it.
  The flake app runs it in a minimal dev shell.

* bench: keep lock profile data out of target

Cluster data now lives in /tmp/etl-lc.XXXXXX so pas... (continued)

48 of 573 new or added lines in 9 files covered. (8.38%)

21 existing lines in 5 files now uncovered.

62331 of 79398 relevant lines covered (78.5%)

49636.06 hits per line

Uncovered Changes

Lines Coverage ∆ File
261
0.0
crates/etl-benchmarks/src/lock_contention/profiler.rs
254
0.0
crates/etl-benchmarks/src/lock_contention/mod.rs
7
79.66
0.35% crates/etl/src/store/both/memory.rs
3
0.0
crates/etl-benchmarks/src/bin/lock_contention.rs

Coverage Regressions

Lines Coverage ∆ File
14
91.08
0.29% crates/etl-destinations/src/ducklake/encoding.rs
3
86.97
0.27% crates/etl-destinations/src/ducklake/core.rs
2
92.68
-0.02% crates/etl-destinations/src/ducklake/client.rs
1
87.27
-0.21% crates/etl-postgres/src/tokio/test_utils.rs
1
83.95
-0.14% crates/etl/src/error.rs
Jobs
ID Job ID Ran Files Coverage
1 44213223634.1 18 Nov 2026 01:14AM UTC 270
78.5
GitHub Action Run
Source Files on build 44213223634
  • Tree
  • List 270
  • Changed 14
  • Source Changed 6
  • Coverage Changed 14
Coverage ∆ File Lines Relevant Covered Missed Hits/Line
  • Back to Repo
  • Github Actions Build #44213223634
  • 57fce02a on github
  • Prev Build on main (#44135990838)
  • Next Build on main (#44265060665)
STATUS · Troubleshooting · Open an Issue · Sales · Support · CAREERS · ENTERPRISE · START FREE TRIAL · SCHEDULE DEMO
ANNOUNCEMENTS · TWITTER · TOS & SLA · Supported CI Services · What's a CI service? · Automated Testing

© 2026 Coveralls, Inc