cmd/pprof: update the vendored tool from upstream #21047
Comments
ALTree
added
the
early-in-cycle
|
google/pprof#169 should fix the file system related issue on Android.
Is it possible to add building and testing of google/pprof at tip to the Go build matrix, "Sub-repositories at tip" section? This would continuously ensure that pprof is Go vendoring ready. |
@bradfitz is it possible/do we want this? |
|
We want that, yes. And everything's possible with a bit of time. We're short on time. But if somebody wants to work on #14594, that'd be great. |
|
@ALTree google/pprof#146 was fixed, can you try updating pprof in Go vendoring area again? |
|
@aalexand nice! I could send a change but it won't be merged now because we're deep deep into the code freeze. I'll send a change as soon as the tree re-open for go1.10 in August. |
|
@ALTree Could you create the change just to see if all the tests pass? |
|
I can do that, but what platforms are you interested in? For example, we have an OSX10.8 builder but not an OSX10.8 trybot. Builders run on master after a commit, trybots can be used to test a change before committing it. So even if I upload a change and run the trybots on it, they won't tell us if the OSX10.8 issue is gone. In fact, with the old failed update in June the trybots said "OK" on my change and only after it was merged I noticed the OSX10.8, OSX10.10 and iPhone failures (by looking at the builders webpage). |
|
I see. Then it doesn't make much sense, no. I didn't know the set of trybot envs is limited. |
|
Change https://golang.org/cl/57370 mentions this issue: |
Update vendored pprof to commit 4fc39a00b6b8c1aad05260f01429ec70e127252c from github.com/google/pprof (2017-11-01). Fixes #19380 Updates #21047 Change-Id: Ib64a94a45209039e5945acbcfa0392790c8ee41e Reviewed-on: https://go-review.googlesource.com/57370 Run-TryBot: Alberto Donizetti <alb.donizetti@gmail.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
|
pprof was vendored and the builders are happy (except the plan9 ones), so I'm closing this issue. I opened #22559 for the plan9 issue. |
Opening this to document a recent (failed) effort to update
pproffrom upstream.The tool was successfully updated from upstream on March 1 (CL 37652). A failed attempt to a final update for go1.9 was made on June 20 (CL 46155). The update broke several builders and was immediately reverted.
The failing builders:
We had an OSX10.8 failure, an OSX10.10 failure, and failures on the
darwin-arm-a1549iosanddarwin-arm64-a1549iosbuilders. All these failures were on the recently introducedTestHttpsInsecuretest. I reported the issue upstream (TestHttpsInsecure failures on certain darwin builders google/pprof#146).Both the android builders were failing (failure1, failure2), again, on
TestHttpsInsecure, but with a different error (related to filesystem permissions, apparently).The
linux-arm-arm5spacemonkeywas failing on theTestHttpsInsecuretest, with the same error we got on the android builders (issues with filesystem permissions).Two of the plan9 builders (
plan9-386andplan9-arm) were failing (failure1, failure2), again, onTestHttpsInsecure, with the same error we got on the darwin builders.To sum it up: all the failures are on the
TestHttpsInsecuretest. Darwin and plan9 builders are failing because the test somehow generates bad/unexpected profiles(?). Alinux-arm5builder and both the android builders are failing with errors related to bad filesystem permission (I suspect the test makes assumptions about write permissions on certain folders that do not hold on some of the builders).The text was updated successfully, but these errors were encountered: