SVNAdmin Load Returns Checksum Mismatch

Question

I'm trying to load a svn dumpfile onto my RHEL server. The dumpfile was loaded from our old Ubuntu server and I used svnadmin create reponame and then svnadmin load --quiet reponame < dumpfilename.

After a short time, I receive the following message:

svnadmin: Checksum mismatch,   
file '/images/lang_cards/large/2001_Lang_Valentines_inside_large.jpg':
expected:  97b1f37b81463ac88a3c2c18abfbb329
actual:  944edbcc7684cb736eca5ed2c0d1c12c

How can I bypass this error, to continue loading the dumpfile?

Thanks!

Solution 1

Possibilities include a corrupt repository, a dumpfile corrupted during transfer, or a hand-edited dumpfile (trying to change paths...?).

If you're trying to change paths, maybe any patching greedily touched revision content, vs just revision entry headings.

If you have no way of re-dumping and trying again (or if doing so has the same problem), try svndumpfilter to exclude that file and any others with problems.

To recover those files, you can very carefully patch the dumpfile to have the correct checksum (Text-content-md5?, but verify the file isn't corrupt), or get them from checkouts/working copies (or extract and reconstruct from the text and deltas in the dump).

Solution 2

Had this problem after filtering my dump with https://github.com/jasperlee108/svndumpfilterIN

Fixed with removing md5 checksums:

sed -i '/Text-copy-source-md5/d' your.dmp

However, there can possible be some consequences...

Comments

[Greg Petersen]

Could you please show us the exactly errors?

[Jehy]

Why silient downvote?

[Sebastian J.]

IMHO it's fair to use your approach if you know what you are doing (i.e. removing a password from the repo): Modify the dump and drop the lines Text-content-md5 and Text-content-sha1. (Obv. you risk having corrupted content)

[TWiStErRob]

Thanks for the tip about the existence of the tool. I tried running svndumptool.py sanitize and got a tiny file with all contents anonymized to "random" hex numbers. When running sanitize -n -u -l -f it re-created a similar dump file, but didn't recalculate neither MD5 nor SHA1 checksums. I tried svndumptool.py check -A -v as well, and it just said "OK", even though svnadmin load fails on it with checksum mismatch.

[TWiStErRob]

Beware: any processing with this tool removed Text-copy-source-md5 and Text-copy-source-sha1 for me.

[TWiStErRob]

I needed to use this variant: sed -re '/^(Text-content-md5|Text-content-sha1|Text-copy-source-md5|‌​Text-copy-source-sha‌​1): /d' <patched.dump >nohash.dump. Note: using </> instead of -i sped things up significantly when running inside docker.