Closed Bug 1101299 Opened 10 years ago Closed 9 years ago

Intermittent jaeger/bug832670.js | Timeout (code -9, args "--no-baseline --no-ion")

Categories

(Core :: JavaScript Engine: JIT, defect)

36 Branch
All
Linux
defect
Not set
normal

Tracking

()

RESOLVED FIXED
mozilla37
Tracking Status
firefox35 --- fixed
firefox36 --- fixed
firefox37 --- fixed
firefox-esr31 --- unaffected

People

(Reporter: KWierso, Assigned: jandem)

Details

(Keywords: intermittent-failure)

Attachments

(1 file)

16:03:21 INFO - TEST-PASS | tests/jit-test/jit-test/tests/jaeger/bug825966.js | Success (code 0, args "--no-baseline --no-ion")
16:03:41 INFO - TEST-PASS | tests/jit-test/jit-test/tests/jaeger/bug832670.js | Success (code 0, args "")
16:03:41 INFO - TEST-PASS | tests/jit-test/jit-test/tests/jaeger/bug832670.js | Success (code 0, args "--ion-eager --ion-offthread-compile=off")
16:03:42 INFO - TEST-PASS | tests/jit-test/jit-test/tests/jaeger/bug832670.js | Success (code 0, args "--ion-eager --ion-offthread-compile=off --ion-check-range-analysis --no-sse3 --no-threads")
16:03:42 INFO - TEST-PASS | tests/jit-test/jit-test/tests/jaeger/bug832670.js | Success (code 0, args "--baseline-eager")
16:03:42 INFO - TEST-PASS | tests/jit-test/jit-test/tests/jaeger/bug832670.js | Success (code 0, args "--baseline-eager --no-fpu")
16:06:12 INFO - TIMEOUT - jaeger/bug832670.js
16:06:12 WARNING - TEST-UNEXPECTED-FAIL | tests/jit-test/jit-test/tests/jaeger/bug832670.js | Timeout (code -9, args "--no-baseline --no-ion")
16:06:12 INFO - INFO exit-status : -9
16:06:12 INFO - INFO timed-out : True
16:06:12 INFO - TEST-PASS | tests/jit-test/jit-test/tests/jaeger/clonefun.js | Success (code 0, args "")
16:06:13 INFO - TEST-PASS | tests/jit-test/jit-test/tests/jaeger/clonefun.js | Success (code 0, args "--ion-eager --ion-offthread-compile=off")
16:06:13 INFO - TEST-PASS | tests/jit-test/jit-test/tests/jaeger/clonefun.js | Success (code 0, args "--ion-eager --ion-offthread-compile=off --ion-check-range-analysis --no-sse3 --no-threads")
16:06:13 INFO - TEST-PASS | tests/jit-test/jit-test/tests/jaeger/clonefun.js | Success (code 0, args "--baseline-eager")
16:06:13 INFO - TEST-PASS | tests/jit-test/jit-test/tests/jaeger/clonefun.js | Success (code 0, args "--baseline-eager --no-fpu")
16:06:13 INFO - TEST-PASS | tests/jit-test/jit-test/tests/jaeger/clonefun.js | Success (code 0, args "--no-baseline --no-ion")
16:06:13 INFO - TEST-PASS | tests/jit-test/jit-test/tests/jaeger/closure-01.js | Success (code 0, args "")
Way too frequent for nobody on the JS team to be investigating.
Flags: needinfo?(jdemooij)
(In reply to Ryan VanderMeulen [:RyanVM UTC-5] from comment #50)
> Way too frequent for nobody on the JS team to be investigating.

I briefly looked into this a few days ago. The test is timing out with --no-baseline --no-ion so it's probably the interpreter being slow. I can't repro this locally though; the test finishes quickly here. Will investigate more soon.
Here's a Try push that dumps the time it takes to run this test:

https://treeherder.mozilla.org/ui/#/jobs?repo=try&revision=759fa5f601b2

Let's see if it's really much slower on Linux 32-bit.
Some numbers for the --no-baseline --no-ion run:

* Linux32 Debug Jit2:      149734 ms (timed out..)
* Linux64 Debug Jit2:       86124 ms
* Linux32 Opt Jit2:             3 ms
* Linux64 Opt Jit2:             2 ms
* Linux32 Debug warnaserr:   1568 ms
* Linux64 Debug warnaserr:   3530 ms

This confirms the problem is worst with the Linux32 Debug Jit2 runs.

First I want to understand the difference between the Jit2 and warnaserr configurations. Both ran on spot instances but warnaserr is way faster (Jit1/Jit2 are always slower I think). Does the warnaserr VM have better specs?
Flags: needinfo?(ryanvm)
(In reply to Jan de Mooij [:jandem] from comment #62)
> Both ran on spot instances but warnaserr is way faster
> (Jit1/Jit2 are always slower I think). Does the warnaserr VM have better
> specs?

I asked philor about this on IRC and he told me the Jit2 job runs on test slaves and warnaserr runs on (beefier) build slaves. Also Amazon 32-bit slaves are slower than the 64-bit ones...
Ah, the test uses gczeal(4), and in the interpreter that means we verify barriers between *each* JSOp. So that could explain (1) why this only affects --no-baseline --no-ion and (2) why Opt builds are super fast.

I want to find out how much RAM this test uses. Maybe the verifier is triggering some serious swapping.
Flags: needinfo?(ryanvm)
Attached patch PatchSplinter Review
Terrence and I discussed this in person during the work week. It's a pretty slow test from the fuzzer that managed to hit a particular bug in JM a long time ago. JM is gone, a lot changed and the test is not really testing anything useful.

In person r=terrence.
Assignee: nobody → jdemooij
Status: NEW → ASSIGNED
Flags: needinfo?(jdemooij)
Attachment #8532099 - Flags: review+
Ryan, can you please land this on Aurora and Beta? :)
Flags: needinfo?(ryanvm)
https://hg.mozilla.org/mozilla-central/rev/ab89014a799a
Status: ASSIGNED → RESOLVED
Closed: 9 years ago
Resolution: --- → FIXED
Target Milestone: --- → mozilla37
You need to log in before you can comment on or make changes to this bug.