kayg/calternal
Watch
1
0
Fork
You've already forked calternal
 0
Code Issues 190 Pull requests 61 Projects Releases Packages Wiki Activity Actions 26

perf: judge the save-soak max against the dev range, plus a sync-overlap check #1050

Closed
kayg wants to merge 0 commits from perf/soak-max-gate into dev
pull from: perf/soak-max-gate
merge into: kayg:dev
Conversation 1 Commits - Files changed - +204
kayg commented 2026-11-10 05:10:33 +00:00
Owner
Copy link

What

A measurement-method change, decided by the orchestrator after the #928 soak diagnosis (comment 89399 there). No product code changes.

  • docs/perf/comparing.md (new) says how a branch is compared with dev on the perf VMs:
    • one perf-turn, holding both locks;
    • interleaved ABAB with dev-vs-dev control pairs;
    • p50/p95/p99 and save_ack as the main gate, each judged against the control.
  • The save-soak max is now judged against the dev arms' range over at least three interleaved arms per side. When a branch arm is above that range, the sync-overlap check decides: it's a regression only if the branch adds a sync inside a save, or one of its own syncs overlaps a save.
  • perf/vm/strace-server.sh runs a server under a seccomp-filtered strace that stops only on fsync, fdatasync and openat, with file names. It forwards SIGTERM so probes can stop it.
  • perf/vm/sync-overlap.py joins a soak arm's per-save log with that trace. It prints per-file sync counts and times, every watched sync and the saves it overlaps, and the sync chain of the slowest saves. It exits 1 when a watched sync overlaps a save.
  • docs/perf/budgets.md and perf/README-perf-vms.md link to the new page.

Why

A soak max is one sample in 900. On the shared perf disk, one disk burst decides which arm gets it. In the four-arm diagnostic on #928, the dev arms had maxima of 89.4 and 86.8 ms and the branch arms 50.3 and 66.7 ms, with 0 branch syncs inside any save. The slowest saves were dev's own four-sync chain stretched by the disk.

Checks

  • python3 -m py_compile perf/vm/sync-overlap.py; bash -n perf/vm/strace-server.sh.
  • Run on the #928 diagnostic data on perf-01: --watch secrets lists 10 witness syncs, none overlapping a save, exit 0. --watch 'notes/meta.db$' reports the 9 in-save meta.db checkpoints, exit 1.
  • No Rust, web or perf-path code changed, so there is no perf run for this PR.

🤖 Generated with Claude Code

## What A measurement-method change, decided by the orchestrator after the #928 soak diagnosis (comment 89399 there). No product code changes. - `docs/perf/comparing.md` (new) says how a branch is compared with dev on the perf VMs: - one `perf-turn`, holding both locks; - interleaved ABAB with dev-vs-dev control pairs; - p50/p95/p99 and `save_ack` as the main gate, each judged against the control. - **The save-soak max** is now judged against the **dev arms' range over at least three interleaved arms per side**. When a branch arm is above that range, the **sync-overlap check** decides: it's a regression only if the branch adds a sync inside a save, or one of its own syncs overlaps a save. - `perf/vm/strace-server.sh` runs a server under a seccomp-filtered strace that stops only on `fsync`, `fdatasync` and `openat`, with file names. It forwards SIGTERM so probes can stop it. - `perf/vm/sync-overlap.py` joins a soak arm's per-save log with that trace. It prints per-file sync counts and times, every watched sync and the saves it overlaps, and the sync chain of the slowest saves. It exits 1 when a watched sync overlaps a save. - `docs/perf/budgets.md` and `perf/README-perf-vms.md` link to the new page. ## Why A soak max is one sample in 900. On the shared perf disk, one disk burst decides which arm gets it. In the four-arm diagnostic on #928, the dev arms had maxima of 89.4 and 86.8 ms and the branch arms 50.3 and 66.7 ms, with 0 branch syncs inside any save. The slowest saves were dev's own four-sync chain stretched by the disk. ## Checks - `python3 -m py_compile perf/vm/sync-overlap.py`; `bash -n perf/vm/strace-server.sh`. - Run on the #928 diagnostic data on perf-01: `--watch secrets` lists 10 witness syncs, none overlapping a save, exit 0. `--watch 'notes/meta.db$'` reports the 9 in-save `meta.db` checkpoints, exit 1. - No Rust, web or perf-path code changed, so there is no perf run for this PR. 🤖 Generated with [Claude Code](https://claude.com/claude-code) 
strace-server.sh runs a server under a seccomp-filtered strace that only
stops on fsync, fdatasync and openat and names every file; it forwards
SIGTERM so probes can stop it. sync-overlap.py joins a soak arm's per-save
log with that trace: per-file sync counts and times, every sync of the
branch's own files with the saves it overlaps, and the sync chain of the
slowest saves. It exits 1 when a watched sync overlaps a save.

Used on #928 to show its witness never syncs inside a save.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
docs(perf): judge the save-soak max against the dev range
Some checks are pending
CI / server (pull_request) Waiting to run
Details
CI / rust (pull_request) Waiting to run
Details
CI / web (pull_request) Waiting to run
Details
Markdown extraction contract / markdown-native-contract (pull_request) Waiting to run
Details
Markdown extraction contract / markdown-editor-contract (pull_request) Waiting to run
Details
0b5d1f11af 
A soak max is one sample in 900; on the shared perf disk one burst decides
which arm has it (#928: dev arms 89.4/86.8 ms, branch arms 50.3/66.7 ms,
nothing branch-only inside any save). comparing.md writes down how a
branch is compared with dev: one perf-turn, both locks, interleaved ABAB
with dev-vs-dev control pairs, p50/p95/p99 and save_ack as the main gate.
The max now passes when every branch arm is inside the dev arms' range over
at least three arms per side; above it, the sync-overlap check decides.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
kayg commented 2026-11-10 09:45:22 +00:00
Author
Owner
Copy link

Merged into dev (dc420d71d); closing.

Merged into dev (dc420d71d); closing.
kayg closed this pull request 2026-11-10 09:45:22 +00:00
Some checks are pending
CI / server (pull_request) Waiting to run
Details
CI / rust (pull_request) Waiting to run
Details
CI / web (pull_request) Waiting to run
Details
Markdown extraction contract / markdown-native-contract (pull_request) Waiting to run
Details
Markdown extraction contract / markdown-editor-contract (pull_request) Waiting to run
Details

Pull request closed

Please reopen this pull request to perform a merge.
Sign in to join this conversation.
Reviewers
No reviewers
Labels
Clear labels   
backend
Rust server

  
design
Mockups, UI and TASTE

  
docs
ADRs, CONTEXT, TASTE

  
idea
Parked idea, never worked until promoted

  
infra
Deploy, CI, VMs

  
perf
Performance and budgets

  
reliability
Data integrity, corruption traps

  
web
Web client

No labels
backend
design
docs
idea
infra
perf
reliability
web
Milestone
Clear milestone
No items
No milestone
Projects
Clear projects
No items
No project
Assignees
Clear assignees
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
kayg/calternal!1050
No description provided.