Skip to content

DO NOT MERGE: measure #1266-#1268 with #1020 on a Rust-changing PR - #1271

Closed
zibi-pplx wants to merge 5 commits into
mainfrom
pr1271
Closed

zibi-pplx wants to merge 5 commits into
mainfrom
pr1271

Conversation

@zibi-pplx

@zibi-pplx zibi-pplx commented Oct 31, 2026 •

Copy link
Copy Markdown
Collaborator

D1 plus #1020's six commits (build the seed early, run the console step
in parallel), with their Tiltfile conflicts against today's main taken
as main's structure plus #1020's change. For measurement only; #1020
itself is unchanged.


Stack created with Sapling. Best reviewed with ReviewStack.

zibi-pplx and others added 5 commits October 31, 2026 19:55
Each Kind smoke shard builds and deploys the Next.js UI during its
bring-up, about 100 seconds, and the bring-up waits for every workload,
so the UI is one of the two steps that end it. No smoke scenario reads
the UI. SPACE_KIND_UI=0 now leaves its image and release out of the
Tiltfile, and the shards set it; a local up.sh still deploys the UI.
…th API surfaces

Each shard runs its ECW scenarios and then its native ones, but each
pass was partitioned on its own, so a shard's two loads did not add up
evenly: by the duration hints, eight such shards would range from 7.3 to
9.9 minutes. A sharded --all pass now partitions both surfaces' eligible
scenarios together, from the same registry and filters, and runs its own
surface's share; every pass computes the same partition, so each
(surface, scenario) pair still runs exactly once. By the hints, eight
shards then carry 7.7 minutes each, which is about the leader-election
brownout's own length, so a ninth would not help. The cost is two more
16-core runners per Kind run.
…allel

The observability install, which Kind CI's shards run while the Rust
producer builds and local bring-ups run from the Tiltfile, did
everything one step at a time: six mirror pulls (about 23 s in CI), six
kind image loads onto every node (about 57 s, Prometheus alone 30 s)
and four Helm releases with --wait (about 60 s). With the producer now
often faster, that two-minute install is on a shard's critical path.

The pulls and the loads now run in parallel, and so do the metrics and
traces releases, each its own chart in its own namespace; the logs and
Grafana releases keep their order after them. The first pull logs in
alone, so no two docker logins write the Docker config at once. Each
parallel command's output is printed in order, and any failure still
fails the install.
The stack plus one comment line in a Rust file, so the Kind producer
builds incrementally from main's state as a typical PR does.
D1 plus #1020's six commits (build the seed early, run the console step
in parallel), with their Tiltfile conflicts against today's main taken
as main's structure plus #1020's change. For measurement only; #1020
itself is unchanged.
@zibi-pplx

Copy link
Copy Markdown
Collaborator Author

Measurement done; the results are in #1267, #1268 and #1266. Closing as planned.

@zibi-pplx zibi-pplx closed this Oct 31, 2026
@zibi-pplx
zibi-pplx deleted the pr1271 branch October 31, 2026 21:48
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants