Skip to content

restore the wtx and sark results deleted with a disabled sibling - #1299

Merged
MDA2AV merged 1 commit into
mainfrom
fix/restore-shared-result-files
Aug 24, 2026
Merged

restore the wtx and sark results deleted with a disabled sibling#1299
MDA2AV merged 1 commit into
mainfrom
fix/restore-shared-result-files

Conversation

@MDA2AV

@MDA2AV MDA2AV commented Aug 24, 2026

Copy link
Copy Markdown
Owner

Results are keyed by display_name, and nine entries share one with a sibling. #1291 deleted whole result files because a single entry writing into them failed — taking every sibling's numbers with it.

wtx

All five wtx entries share display_name: wtx, so they publish into one wtx.json. wtx-grpc and wtx-grpc-tls failed their gRPC checks, and the deletion removed the file:

profile keys owned by still enabled?
baseline-h2, baseline-h2c, json-h2c 7 wtx-http2, wtx-http2-tls yes
echo-ws, echo-ws-limited, echo-ws-pipeline 8 wtx-ws yes
unary-grpc, unary-grpc-tls 4 wtx-grpc, wtx-grpc-tls no — correctly disabled

15 of the 19 keys belonged to entries that were never failing. WTX has been showing nothing on the board since.

sark

Same shape: sark was disabled, and gateway-64 / gateway-h3 went with it — 4 keys owned by sark-gateway and sark-production, both still enabled.

zix — deliberately not restored

Its file was deleted the same way, but none of the 23 keys belonged to zix-http2, zix-grpc, zix-ws or zix-http3; every one was published by the disabled zix entry. An empty file is worse than no file.

What this fixes and what it does not

Both files return with only the keys the still-enabled entries own. The profiles the disabled entries published stay gone — that part of #1291 was correct.

The six families sharing a display_name where every entry is enabled — Cardigan, actix, aspnet-minimal, mq-bridge, quarkus-jvm, swoole — were never touched and are unaffected.

Root cause

The deletion logic mapped entry → display_name slug → delete that file. That is right when one entry owns a file and wrong when five share it. Anything deleting results by display_name has to check whether a sibling still publishes into it.

wtx is back in 15 profile views, sark in 4, badge parity holds at 530.

Results are keyed by display_name, and several entries share one. #1291
deleted whole result files because a single entry that writes into them
failed, taking every sibling's numbers with it:

  wtx     wtx-grpc and wtx-grpc-tls failed their gRPC checks, and the
          deletion took baseline-h2, baseline-h2c, json-h2c and all three
          echo-ws profiles with them -- 15 keys belonging to wtx-http2,
          wtx-http2-tls and wtx-ws, all of which are still enabled and were
          never failing. WTX has been showing nothing on the board since.
  sark    the same shape:  was disabled, and gateway-64 and gateway-h3
          went with it -- 4 keys owned by sark-gateway and sark-production.

Both files come back with only the keys the still-enabled entries own; the
profiles the disabled entries published are gone, which is what the original
change was for.

zix is not restored. Its file was deleted the same way, but none of the 23
keys in it belonged to zix-http2, zix-grpc, zix-ws or zix-http3 -- every one
was published by the disabled  entry. An empty file is worse than no
file, so it stays deleted.

The five families that share a display_name and are wholly enabled --
Cardigan, actix, aspnet-minimal, mq-bridge, quarkus-jvm, swoole -- were
never touched by those changes and are unaffected.

wtx is back in 15 profile views, sark in 4, and badge parity holds at 530.
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.

1 participant