Summary: | LayoutTests intermittently failing to run due to issues starting Web Platform Test server | ||||||||
---|---|---|---|---|---|---|---|---|---|
Product: | WebKit | Reporter: | Ryan Haddad <ryanhaddad> | ||||||
Component: | Tools / Tests | Assignee: | youenn fablet <youennf> | ||||||
Status: | RESOLVED FIXED | ||||||||
Severity: | Normal | CC: | ap, cdumez, commit-queue, glenn, gsnedders, lforschler, webkit-bug-importer, youennf | ||||||
Priority: | P2 | Keywords: | InRadar | ||||||
Version: | WebKit Nightly Build | ||||||||
Hardware: | Unspecified | ||||||||
OS: | Unspecified | ||||||||
See Also: |
https://bugs.webkit.org/show_bug.cgi?id=160422 https://bugs.webkit.org/show_bug.cgi?id=164822 |
||||||||
Attachments: |
|
Description
Ryan Haddad
2016-06-27 13:52:43 PDT
Not sure if this could be related to <https://trac.webkit.org/changeset/202471> I am not sure either but it also happens on iOS sim EWS bots, this is rather annoying :( It seems the web socket server is not able to start properly. https://trac.webkit.org/changeset/202471 is not supposed to change the web socket server implementation so I am not sure where that comes from. Do you know where I can find the full wpt_wk log? Created attachment 282223 [details]
Removing ws port from config
You can get to the logs by going to test results, and removing "results.html" from the URL, e.g. <https://build.webkit.org/results/Apple%20Yosemite%20Release%20WK1%20(Tests)/r202508%20(15787)/>. The proposed patch doesn't seem related to this error. This is getting very frequent, and a real problem. Youenn, could you take another look? E.g. in this patch, two EWS bubbles are orange at once because the bots can't start wtpserve: <https://bugs.webkit.org/show_bug.cgi?id=159287>. (In reply to comment #7) > E.g. in this patch, two EWS bubbles are orange at once because the bots > can't start wtpserve: <https://bugs.webkit.org/show_bug.cgi?id=159287>. From the error log, it seems bots have difficulties generating certificates from openssl. We can try regenerating the certificates and I will prepare a patch accordingly. But there may be an underlying issue... Created attachment 282416 [details]
Patch
(In reply to comment #9) > Created attachment 282416 [details] > Patch Hopefully this should work. I think there may be a bug in some of openssl wpt scripts. The culprit might be https://github.com/w3c/wpt-tools/commit/b388d6b14df8acc3fad070c74b435cd7aff056d1 Raised issue here: https://github.com/w3c/wpt-tools/issues/86 (In reply to comment #12) > Raised issue here: https://github.com/w3c/wpt-tools/issues/86 OK, the issue is already fixed in w3c repo. Instead of this patch, we could reimport the latest wpt repo. Comment on attachment 282416 [details]
Patch
I don't quite understand this change, nor whether it's practical to reimport quickly.
Does this prevent re-generating the certificates on every run, and thus sidestep the problem in automatic generation?
Either way, r=me, as we need to resolve this ASAP.
> I don't quite understand this change, nor whether it's practical to reimport > quickly. Reimporting would fix the bug without us having to change our current setup. > Does this prevent re-generating the certificates on every run, and thus > sidestep the problem in automatic generation? You are correct. I don't think serial is used in wpt except for generating certificates when launching servers, hence why the problem should not appear anymore. Comment on attachment 282416 [details] Patch Clearing flags on attachment: 282416 Committed r202681: <http://trac.webkit.org/changeset/202681> All reviewed patches have been landed. Closing bug. Comment on attachment 282416 [details]
Patch
After this change, LayoutTests/imported/w3c/resources/_wpt_certs/* keep getting overridden locally and showing up in my local diff / patches. This is very annoying.
Ah, it might be due to recent changes to wpt python scripts. We can go back to the previous strategy since not-closed-handle bug is now fixed. Seeing something similar here again: https://build.webkit.org/builders/Apple%20Sierra%20Release%20WK2%20%28Tests%29/builds/1463/steps/layout-test/logs/stdio I looked at one of the bots hitting this, and it has lots of connections that are not properly closed ("sunwebadmin" is port 8800): tcp4 0 0 localhost.sunwebadmin localhost.62045 TIME_WAIT tcp4 0 0 localhost.sunwebadmin localhost.62049 TIME_WAIT tcp4 0 0 localhost.sunwebadmin localhost.62048 TIME_WAIT tcp4 0 0 localhost.sunwebadmin localhost.61980 TIME_WAIT tcp4 0 0 localhost.sunwebadmin localhost.61984 TIME_WAIT tcp4 0 0 localhost.sunwebadmin localhost.61982 TIME_WAIT tcp4 0 0 localhost.sunwebadmin localhost.61986 TIME_WAIT tcp4 0 0 localhost.sunwebadmin localhost.61988 TIME_WAIT tcp4 0 0 localhost.sunwebadmin localhost.61990 TIME_WAIT tcp4 0 0 localhost.sunwebadmin localhost.62051 TIME_WAIT tcp4 0 0 localhost.sunwebadmin localhost.62055 TIME_WAIT tcp4 0 0 localhost.sunwebadmin localhost.62054 TIME_WAIT tcp4 0 0 localhost.sunwebadmin localhost.62057 TIME_WAIT tcp4 0 0 localhost.sunwebadmin localhost.62059 TIME_WAIT tcp4 0 0 localhost.sunwebadmin localhost.62061 TIME_WAIT tcp4 0 0 localhost.sunwebadmin localhost.62063 TIME_WAIT tcp4 0 0 localhost.sunwebadmin localhost.62067 TIME_WAIT tcp4 0 0 localhost.sunwebadmin localhost.62066 TIME_WAIT tcp4 0 0 localhost.sunwebadmin localhost.62069 TIME_WAIT tcp4 0 0 localhost.sunwebadmin localhost.62073 TIME_WAIT tcp4 0 0 localhost.sunwebadmin localhost.62072 TIME_WAIT tcp4 0 0 localhost.sunwebadmin localhost.62075 TIME_WAIT tcp4 0 0 localhost.sunwebadmin localhost.62079 TIME_WAIT tcp4 0 0 localhost.sunwebadmin localhost.62078 TIME_WAIT tcp4 0 0 localhost.sunwebadmin localhost.62081 TIME_WAIT tcp4 0 0 localhost.sunwebadmin localhost.62086 TIME_WAIT tcp4 0 0 localhost.sunwebadmin localhost.62087 TIME_WAIT tcp4 0 0 localhost.sunwebadmin localhost.62085 TIME_WAIT tcp4 0 0 localhost.sunwebadmin localhost.62093 TIME_WAIT tcp4 0 0 localhost.sunwebadmin localhost.62097 TIME_WAIT tcp4 0 0 localhost.sunwebadmin localhost.62096 TIME_WAIT tcp4 0 0 localhost.sunwebadmin localhost.62099 TIME_WAIT tcp4 0 0 localhost.sunwebadmin localhost.62102 TIME_WAIT tcp4 0 0 localhost.sunwebadmin localhost.62103 TIME_WAIT tcp4 0 0 localhost.sunwebadmin localhost.62105 TIME_WAIT tcp4 0 0 localhost.sunwebadmin localhost.62108 TIME_WAIT tcp4 0 0 localhost.sunwebadmin localhost.62109 TIME_WAIT tcp4 0 0 localhost.sunwebadmin localhost.62111 TIME_WAIT tcp4 0 0 localhost.sunwebadmin localhost.62114 TIME_WAIT tcp4 0 0 localhost.sunwebadmin localhost.62115 TIME_WAIT tcp4 0 0 localhost.sunwebadmin localhost.62117 TIME_WAIT tcp4 0 0 localhost.sunwebadmin localhost.62121 TIME_WAIT tcp4 0 0 localhost.sunwebadmin localhost.62120 TIME_WAIT tcp4 0 0 localhost.sunwebadmin localhost.62123 TIME_WAIT tcp4 0 0 localhost.sunwebadmin localhost.62127 TIME_WAIT tcp4 0 0 localhost.sunwebadmin localhost.62126 TIME_WAIT tcp4 0 0 localhost.sunwebadmin localhost.62129 TIME_WAIT tcp4 0 0 localhost.sunwebadmin localhost.62133 TIME_WAIT tcp4 0 0 localhost.sunwebadmin localhost.62132 TIME_WAIT tcp4 0 0 localhost.sunwebadmin localhost.62135 TIME_WAIT (over 4000 connections in TIMED_WAIT on that bot) (In reply to comment #22) > (over 4000 connections in TIMED_WAIT on that bot) I run 50 times LayoutTests/imported/w3c/web-platform-tests/fetch/api/request/request-structure.html which has two sub-resources. We end up with 151 TIME_WAIT sockets. Looking at the server, it seems to support HTTP keep-alive and socket reuse (In reply to comment #23) > (In reply to comment #22) > > (over 4000 connections in TIMED_WAIT on that bot) > > I run 50 times > LayoutTests/imported/w3c/web-platform-tests/fetch/api/request/request- > structure.html which has two sub-resources. > We end up with 151 TIME_WAIT sockets. > > Looking at the server, it seems to support HTTP keep-alive and socket reuse But most responses do not have content-length headers |