changed LP name: kamion -> cjwatson
add comment about buggy svn listdir
|Deletions are marked like this.||Additions are marked like this.|
|Line 46:||Line 46:|
|* [https://launchpad.net/people/cjwatson cjwatson]: [https://launchpad.net/products/debconf/head debconf/head] [[BR]] svn://svn.debian.org/debconf/trunk/src/debconf [[BR]] [[Date(2006-11-28T11:51:08Z)]]||It could be that some of these failures are caused by another specific bug. From a discussion with Jelmer:
> > > * Don't rely on listdir() kind results. Fixes imports of several
> > > repositories. (#56647)
> > I'm curious about this bug. Can you tell me more?
> It turned out that the .kind member of dirents returned by
> svn.ra.get_dir2() is (semi-?)random when used against older Subversion
> servers. It would occasionally be 1 (svn.core.svn_node_file) even though
> the entry in question was a directory. This took me a while to track
> down. Instead, I've switched to always trying to open something as a
> directory and catching the appropriate exception that indicates it's a
That's very interesting. A number of Launchpad imports fail in a not-deterministic way with "failed REPORT". Currently I just classify those failures as "unreliable server", but I was puzzled by a pattern I noticed: the failures did not happen in a really random way.
It might be the same bug.
|Line 49:||Line 66:|
* nobody: [https://launchpad.net/products/reconstructor/trunk reconstructor/trunk] [[BR]] http://reconstructor.googlecode.com/svn/trunk [[BR]] [[Date(2006-11-28T11:51:08Z)]]
This page tracks requests for actions from the vcs-imports operator.
To request an import, please:
- Create a product in Launchpad and set the source details to the CVS or Subversion repository you wish to have imported. In particular
- Add a request on this page, with the name of your launchpad account so I can find your email.
Note that we only do imports of MAIN branches on CVS (the form will have to be updated to remove the "branch" field) and trunk branches on Subversion.
We will handle your request and keep you informed of progress or issues. You can contact the vcs-imports operator: ddaa in #launchpad on irc.freenode.net.
Template to request an import:
[https://launchpad.net/people/<launchpad-user> <launchpad-user>]: [https://launchpad.net/products/<launchpad-product>/<series> <product>/<series>] BR <location> BR @DATE@
Imports that need intervention from the vcs-imports operator.
Imports being processed.
Imports that are currently failing, and for which no fix is currently being worked on.
[svn] wrong repository root
The current svn import code determines the root of the repository by walking up the URL until svn ls fails. This was necessary because older versions of libsvn_client did not expose the repository root information. Newer libsvn_client allow to determine this information directly. In cases where there are nested svn repositories, the old logic is wrong and that cause the change filtering (ignoring change outside of the branch being imported) to ignore legitimate changes.
[svn] unreliable server
The current svn import code makes many connections and does not retry when one fails. If the server is in the least unreliable (one failure every few thousand attempts to create a session) that will cause the import to fail.
It could be that some of these failures are caused by another specific bug. From a discussion with Jelmer:
> > > * Don't rely on listdir() kind results. Fixes imports of several > > > repositories. (#56647) > > I'm curious about this bug. Can you tell me more? > It turned out that the .kind member of dirents returned by > svn.ra.get_dir2() is (semi-?)random when used against older Subversion > servers. It would occasionally be 1 (svn.core.svn_node_file) even though > the entry in question was a directory. This took me a while to track > down. Instead, I've switched to always trying to open something as a > directory and catching the appropriate exception that indicates it's a > file. That's very interesting. A number of Launchpad imports fail in a not-deterministic way with "failed REPORT". Currently I just classify those failures as "unreliable server", but I was puzzled by a pattern I noticed: the failures did not happen in a really random way. It might be the same bug.
[https://launchpad.net/people/cjwatson cjwatson]: [https://launchpad.net/products/debian-installer/main debian-installer/main] BR svn://svn.debian.org/d-i/trunk/installer BR Date(2006-11-28T11:51:08Z)
nobody: [https://launchpad.net/products/xorg-driver-synaptics/main xorg-driver-synaptics-main] BR https://oops.kerneljanitors.org/repos/synaptic/trunk BR Date(2006-12-06T10:40:09Z)
[svn] missing uri-encoding
cscvs uses the svn_client API to libsvn to retrieve data (on Date(2006-11-09T14:31:29Z)), but does not correctly uri-encode locations. So file names that contain characters that must be escaped cause import failures.
[cvs] failed to open the modules file
cscvs needs to read the modules file to ignore configs (aka externals, nested trees).
$ cvs -d :pserver:guest:firstname.lastname@example.org:/cvs co -c cvs [checkout aborted]: failed to open the modules file
nobody: [https://launchpad.net/products/clientcookie/trunk clientcookie/trunk] BR http://codespeak.net/svn/wwwsearch/ClientCookie/trunk BR Date(2006-12-11T14:44:59Z)
[cvs] file created by merge
Anchor(epiphany-browser)[https://launchpad.net/people/seb128 seb128]: https://launchpad.net/products/epiphany-browser/main BR :pserver:email@example.com:/cvs/gnome epiphany BR [Date(2006-10-16T14:54:17Z)]
- The cscvs log (with changesets) first mentions epiphany-toolbar at changeset MAIN.67, it refers to revision 1.2 of this file and records it as a 'M' (modification), which fails to apply since the file is not present in the imported tree at this point.
Relevant fragment of the cvs log:
$ cvs rlog epiphany/data\/ui/epiphany-toolbar.xml.in RCS file: /cvs/gnome/epiphany/data/ui/Attic/epiphany-toolbar.xml.in,v head: 1.17 branch: locks: strict access list: symbolic names: [...] eog-menu-api: 220.127.116.11 keyword substitution: kv total revisions: 19; selected revisions: 19 description: ---------------------------- [...] ---------------------------- revision 1.2 date: 2003/01/20 18:57:15; author: mpeseng; state: Exp; lines: +15 -0 2003-01-20 Marco Pesenti Gritti <firstname.lastname@example.org> * Merge eog-menu-api branch ---------------------------- revision 1.1 date: 2003/01/20 17:53:25; author: mpeseng; state: dead; branches: 1.1.2; file epiphany-toolbar.xml.in was initially added on branch eog-menu-api. ---------------------------- revision 18.104.22.168 date: 2003/01/20 17:53:25; author: mpeseng; state: Exp; lines: +15 -0 implement context menus ---------------------------- [...]
[svn] Cannot import whole repository
The svn import code fails at processing imports at the root of a svn repository. Some repository are set up that way, with the actual root containing the source code. That's broken, but people actually do this.
[cvs] None branch in findLastFileRevision
While creating changesets: assertion error, branchName is None in findLastFileRevision CVS/CacheGenerator.py:160
It looks like it means something is really busted with filler revisions and default branches.
File name not supported by bzr
[cvs] Absurd filler revision
on changeset MAIN.4167: BR CRITICAL: requested checkout of reporoot '/cvsroot/xrms', file 'xrms/include/adodb/pear/Attic/readme.Auth.txt', revision '22.214.171.124' BR CRITICAL: log of checkout responses follow. BR CRITICAL: "E cvs checkout: cannot find module BR `xrms/include/adodb/pear/Attic/readme.Auth.txt' - ignored\n" BR CRITICAL: 'error \n'
cscvs creates a filler changeset 1.1->126.96.36.199 for include/adodb/pear/readme.Auth.txt, but there is no revision 188.8.131.52 for that file. The cvs rlog shows revision 1.1 (branches: 1.1.1) used in changeset MAIN.915, revision 184.108.40.206, not used by cscvs. The cvs log header for that file reads "head: 1.1" and "branch: 1.1.1".
(15:28:25) lifeless: I'd do a search in the file revision table for that file, all revisions, see what cscvs is thinking happened to it (15:29:45) ddaa: you suggest it might be a problem in the log parser and checking the catalog contents would help narrowing the cause of the problem? (15:33:11) lifeless: something is inventing the .2 (15:33:22) lifeless: if its in the catalog then the catalog generator can be checked (15:33:42) lifeless: if its not in the catalog then the application-of-changesets can be checked (15:33:45) lifeless: it will narrow it down (15:34:03) lifeless: As its a very small rlog I'd make a test case from it
on changeset MAIN.394 BR CRITICAL: requested checkout of reporoot '/cvs/gnome', file 'gtetrinet/src/Attic/tetris.h', revision '220.127.116.11' BR CRITICAL: log of checkout responses follow. BR CRITICAL: "E cvs server: cannot find module `gtetrinet/src/Attic/tetris.h' - ignored\n" BR CRITICAL: 'error \n'
cscvs creates a filler changeset 1.1->18.104.22.168 for gtetrinet/src/tetris.h but the log of that file only has revisions 1.1, 22.214.171.124, 126.96.36.199.2.2, 188.8.131.52.2.1.
[cvs] Checkout gave no file data
WARNING: N changeset MAIN.745 CRITICAL: checkout did not give file data and did not indicate file was removed. CRITICAL: requested checkout of reporoot '/cvs/gnome', file 'tomboy/www/img/intro.gif', revision '1.1' CRITICAL: log of checkout responses follow. CRITICAL: 'ok\n'
The cscvs log for this changeset:
ChangeSet: 745 Author: orph Date: 2004/09/18 08:52:21 Branch: MAIN CVSDefaultFiller Revisions: F www/img/intro.gif [1.1->1.1] +0-0 Filler changeset 1.1->1.1
[svn] repository password protected
- On full-tree import of initial revision
WARNING: change 161 WARNING: change 166 Traceback (most recent call last): File "/home/importd/dists/launchpad/lib/importd/CommandLineRunner.py", line 67, in ? main() File "/home/importd/dists/launchpad/lib/importd/CommandLineRunner.py", line 64, in main doer.runMethod(method, dirname) File "/home/importd/dists/launchpad/lib/importd/CommandLineRunner.py", line 57, in runMethod getattr(self._job, method)(logger=self.logger, dir=dirname) File "/srv/importd.ubuntu.com/dists/launchpad/lib/importd/Job.py", line 188, in runJob strategy(self, dir, logger) File "/srv/importd.ubuntu.com/dists/launchpad/lib/importd/JobStrategy.py", line 100, in Import self.runtobaz("-SI", "%s.1" % aJob.branchfrom, target_path, logger) File "/srv/importd.ubuntu.com/dists/launchpad/lib/importd/JobStrategy.py", line 84, in runtobaz totla.totla(config, logger, config.args, self.sourceTree()) File "/srv/importd.ubuntu.com/dists/launchpad/lib/cscvs/cmds/totla.py", line 262, in totla apply_changeset(source_tree, logger, apply, revision, target_tree, revisions) File "/srv/importd.ubuntu.com/dists/launchpad/lib/cscvs/cmds/totla.py", line 130, in apply_changeset fileChange.apply(logger, target_tree, prune) File "/srv/importd.ubuntu.com/dists/launchpad/lib/svn_oo/PresentFile.py", line 46, in apply self.url(), self.revision) File "/srv/importd.ubuntu.com/dists/launchpad/lib/svn_oo/util.py", line 116, in writeFile revision=pysvn_revision, peg_revision=pysvn_revision) pysvn._pysvn.ClientError: Code: 00020f58
Failed to fork, memory error
Some large imports fail to os.fork() (when spawning gpg to sign a revision) with "OSError: [Errno 12] Cannot allocate memory". It's not clear why that happen, since Linux should overcommit memory up to 50x (the value of /proc/sys/vm/overcommit_ratio) and allow large processes like cscvs to spawn small processes like gpg.
- On changeset changeset 5343
- On changeset MAIN.15819
- On changeset 7
Fix in progress
Failures for which a fix is currently being worked on.