[svn.haxx.se] · SVN Dev · SVN Users · SVN Org · TSVN Dev · TSVN Users · Subclipse Dev · Subclipse Users · this month's index

Re: corrupt revision, "Reading one svndiff window read beyond the end of the representation"

From: Daniel Shahaf <d.s_at_daniel.shahaf.name>
Date: Wed, 4 Aug 2010 00:07:34 +0300

Justin Georgeson wrote on Tue, Aug 03, 2010 at 13:37:22 -0500:
> The new test repo up to r39244 was created and the merge was successfully committed. Checking the rev file in that new repo it looks ok. But when I put the new rev file into an existing repo it is still corrupt, although for different reasons now. I tried fsfsverify.py on this new rev file just in case, same error.
> [svnadmin_at_hourdcm1 /tmp]$ svnadmin verify -r 39245 /u1/repos/prowess
> svnadmin: Corrupt node-revision 'g-17515.0-38388.r38555/6646'
> svnadmin: Found malformed header in revision file
> That string comes from this entry
> K 19
> SmartInputTool.java
> V 32
> file g-17515.0-38388.r38555/6646

This is part of the representation of the directory node which contains

> Which is the entry just prior to the entry for the file that was modified in this merge. If I change the entry to match that of the original rev file then (change the trailing /6646 to /6649) then the verify fails with a checksum mismatch. Interestingly I don't see the 'actual' checksum anywhere in the file, but I do see the expected.

The 6646|6649 is a byte offset into the revision file revs/38/38555.

Presumably only one of the two offsets is correct here...

> [svnadmin_at_hourdcm1 /tmp]$ svnadmin verify -r 39245 /u1/repos/prowess
> svnadmin: Checksum mismatch while reading representation:
> expected: 2acda48d7b91e8f07aff6270fdcb9e7b
> actual: 7d22c19004b23609f3a460fb9adbed96
> [svnadmin_at_hourdcm1 /tmp]$ grep 2acda48d7b91e8f07aff6270fdcb9e7b /u1/repos/prowess/db/revs/39/39245
> text: 39245 605 1191 1191 2acda48d7b91e8f07aff6270fdcb9e7b

This tells you that if you seek to the 605th byte of revs/39/39245, you'll
find a representation of length 1191 bytes (after stripping "DELTA"/"PLAIN"
headers and "ENDREP" trailers) to create a file of length 1191 bytes. The
file which the representation expands to shall have an md5 checksum of

> [svnadmin_at_hourdcm1 /tmp]$ grep 7d22c19004b23609f3a460fb9adbed96 /u1/repos/prowess/db/revs/39/39245
> [svnadmin_at_hourdcm1 /tmp]$

Well, now you know what these numbers mean.

HTH; I have to sleep now, so I'll check back on this thread tomorrow

> -----Original Message-----
> From: Justin Georgeson
> Sent: Tuesday, August 03, 2010 10:03 AM
> To: 'Daniel Shahaf'
> Cc: users_at_subversion.apache.org
> Subject: RE: corrupt revision, "Reading one svndiff window read beyond the end of the representation"
> Regarding reproducibility, that's what I was going for with #3. I found another thread, http://svn.haxx.se/users/archive-2009-06/0723.shtml, concluding this error is due to fsfsverify not being current with the latest format, so I'll give the svndump to new repo and redo revision in new repo a try.
> -----Original Message-----
> From: Daniel Shahaf [mailto:d.s_at_daniel.shahaf.name]
> Sent: Tuesday, August 03, 2010 9:30 AM
> To: Justin Georgeson
> Cc: users_at_subversion.apache.org
> Subject: Re: corrupt revision, "Reading one svndiff window read beyond the end of the representation"
> Justin Georgeson wrote on Mon, Aug 02, 2010 at 17:39:49 -0500:
> > I have a repo with >39k revisions. Last week, r39245 was committed, a merge of a single file from trunk to branch. It is the HEAD revision of that file on that branch. Turns out this revision is corrupt
> >
> > [svnadmin_at_hourdcm3 ~]$ svnadmin verify -r 39245 /repos/prowess
> > svnadmin: Reading one svndiff window read beyond the end of the representation
> >
> Is this reproducible? i.e., if you re-commit r39245 (on top of, say, an
> svnsync/backup repository at r39244), does it become corrupted again?
> > I've searched from r30000 to HEAD in this repo and that's the only rev that fails the verify. All our backup copies have the same issue too. I'm wondering what our options for recovery are. Some suggestions we have come up with internally are:
> >
> > 1. Developer still has sandbox which reports the parent folder as updated, so have him 'svn cat' the previous version and commit that, then re-commit the changes from the corrupt revision
> > 2. 'svn rm' the file from the server and re-add it (losing ancestry)
> > 3. Some combination of svndump up to that revision, import to new repo, redo that merge in new repo, overwrite the revision file with new one
> > 4. delete revision file (seems like bad idea)
> > 5. svn dump up to corrupt revision and everything after bad revision, merge dumps, create new repo, redo merge
> >
> Don't do #4.
> #5 sounds reasonable. You have to restitch history in some way now.
> > Is there something else we missed? Which of these seems like the safest/easiest?
> >
> > ----------------------------------------------------------------------
> > This e-mail, including any attached files, may contain confidential and privileged information for the sole use of the intended recipient. Any review, use, distribution, or disclosure by others is strictly prohibited. If you are not the intended recipient (or authorized to receive information for the intended recipient), please contact the sender by reply e-mail and delete all copies of this message.
Received on 2010-08-03 23:09:50 CEST

This is an archived mail posted to the Subversion Users mailing list.

This site is subject to the Apache Privacy Policy and the Apache Public Forum Archive Policy.