kz930 commented on PR #7207:
URL: https://github.com/apache/texera/pull/7207#issuecomment-5245502913

   Thanks — all of it is addressed in the two commits above, and each of the 
three real defects was reproduced before being fixed.
   
   An interrupted caller now discards its worker rather than losing the slot: a 
new case fills every slot with an unanswered job, interrupts the callers, then 
asks the pool to serve that many jobs again, and without the fix it spins at 
the cap until the bound runs out. Starting a worker is now part of the pool's 
`WorkerDiedException` contract rather than something one call site has to name, 
so a bare `IOException` no longer walks past a caller's fallback and the 
runtime driver gets the same protection when it converts. And a first line that 
is not the protocol counts as not ready, with the line quoted in the message — 
stderr is empty in exactly that case — and the interpreter killed, since 
nothing else would reap a worker rejected before it joined the set; the fixture 
grew a mode that writes such a line, which raised `JsonParseException` before. 
On the silent half of that concern: the `py_compile` check now counts the 
descriptors the pool could not serve into its summary. Pointed
  at an interpreter that does not exist, all 117 fall back, all 117 still pass, 
and the summary reports 117 fallbacks instead of reading as a green pooled run.
   
   `-P4` is dropped, together with the shared argument that existed to carry it 
and the `maxWorkers` rationale it rested on — you are right that the cap is per 
sub-pool and env-overridable — and it comes back with the first tagged test 
that pools workers. Whether the fixture answers now travels in the request 
rather than in a launch argument, so the case for serving jobs after a discard 
acts on the sub-pool the discard happened in rather than a fresh one. 
`TestFilters` documents `envVar` and `tag`, and the two comment corrections are 
in as written.
   
   `WorkflowOperator/test` is green at 2163 tests, the pool spec at 8, 
`testOptions` in both modules is unchanged apart from the dropped parameter, 
and scalafmt and scalafix are clean.
   


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to