The WinCairo bot started to fail again with the following error: svn: E120106: ra_serf: The server sent a truncated HTTP response body. First failing build: https://build.webkit.org/builders/WinCairo%2064-Bit%20Release/builds/44298 It stucked in an infinte loop again: - try to checkout the WebKit repository - fail with this error - rm -rf <repository> I noticed this error when this bot was resurrected, and it happened again.
Alex -- it looks like someone updated subversion to 1.8. I've found that this version fails (with this error); we have to stay with subversion 1.7.
I fixed it manually and left it with svn 1.8. I could change the svn version if it becomes a common problem. If it did svn cleanup instead of rm -rf, it would have fixed the problem itself.
(In reply to comment #2) > I fixed it manually and left it with svn 1.8. I could change the svn > version if it becomes a common problem. If it did svn cleanup instead of rm > -rf, it would have fixed the problem itself. rm -rf on SVN error is buildbot internal, unfortunately you can't change it easily. But we used a workaround for it on Linux long time ago because of internet connection issues. It should work on Windows too. The trick is an SVN wrapper script, which tries to run the original command first. If it fails, it calls svn cleanup and then retry the original command. #!/bin/bash SVN_BIN=/usr/bin/svn if [ "x$1" != "xlog" -a "x$1" != "xinfo" ] ; then ($SVN_BIN $*) || ($SVN_BIN cleanup &>/dev/null && $SVN_BIN $*) else $SVN_BIN $* fi Of course it isn't an ultimate solution, because what if the command fails more than once. But it was unlikely, one cleanup was enough for us in 99% of the cases. One important thing: the original output must be the same, because the buildbot parses it and doesn't like extra output generated by cleanup or echoing some debug information. I can't remember exactly why, but svn info and log didn't like this cleanup at all.
I'll just keep my eye on it. If it becomes a problem, I'll install the old version of svn.
I know it isn't related to svn ... but now JSC tests started to fail again after 20 mins timeout without doing anything and binding tests with the strange differences. Something is weird happening there.