attic check --repository-only reports "Error reading segment ###" at different locations

[tgharold]

This is with Attic 0.13 - the "error reading" point is different across multiple runs when checking a repository that is accessed and backed up via SSH. Multiple runs produce different error locations.

attic check --repository-only ssh://example.com/backup/beta/athens/attics/smb-software.attic

Starting repository check... Error reading segment 6322 Index object count mismatch. 967476 != 967402 attic: Exiting with failure status due to previous errors

Running it again produces the output:

Starting repository check... Error reading segment 2531 Error reading segment 4117 Index object count mismatch. 967476 != 967319 attic: Exiting with failure status due to previous errors

On the third run, it reports a different segment number as the culprit:

Starting repository check... Error reading segment 7980 Index object count mismatch. 967476 != 967407 attic: Exiting with failure status due to previous errors

Running it a 4th time, gives yet another different result:

Starting repository check... Error reading segment 300 Error reading segment 1263 Index object count mismatch. 967476 != 967291 attic: Exiting with failure status due to previous errors

Running the check locally on the remote system results in a different result:

attic check --repository-only /backup/beta/athens/attics/smb-software.attic

Starting repository check... Error reading segment 4985 Index object count mismatch. 967476 != 967404 attic: Exiting with failure status due to previous errors

Approximate details about this particular repository is:

Initializing cache...

Archive name: daily-20150113-1800 Archive fingerprint: c80f6ede0b77be9e63f5cfb78bf5031d50d671e333b4ce265c0705a4d45d7061 Start time: Tue Jan 13 18:00:05 2015 End time: Tue Jan 13 22:25:17 2015 Duration: 4 hours 25 minutes 12.22 seconds Number of files: 157115

                   Original size      Compressed size    Deduplicated size

This archive: 72.53 GB 65.00 GB 52.44 GB

All archives: 72.53 GB 65.00 GB 52.44 GB

So it's only a modest sized backup. The "index.#####" file is only 81MB. There are about 9900 segment files under the data/0 directory in the repository.

[ThomasWaldmann]

@tgharold can you reproduce with 0.14 also?

[tgharold]

I'll have to give it a try. What I found was re-running "attic --check --repository-only" would constantly give errors at different locations as shown. It wasn't until I ran "attic --check" without the "--repository-only" flag that attic was able to fix the repository and find the broken chunk. Once the chunk was fixed, things worked fine.

[narkisr]

This happens to me in 0.1.4:

Starting repository check... Error reading segment 13478 Error reading segment 14429 Index object count mismatch. 1279463 != 1279310 attic: Exiting with failure status due to previous errors

[ThomasWaldmann]

Is the corruption maybe related to this?: msgpack/msgpack-python#124

@tgharold @narkisr : Please state the msgpack version you used in your test and whether it used the compiled C version of msgpack or the pure-python version of msgpack.

[whiteadam]

I think I might be having this issue also. If it helps, i'm using the pre-compiled version 0.15 with msgpack._packer.so

Sorry if that's not the right info, I'm not a big C or Python guy :)

Starting repository check...
Error reading segment 0
Error reading segment 1
Error reading segment 2
Error reading segment 3
Error reading segment 4
Error reading segment 5
Error reading segment 6
Error reading segment 7
Error reading segment 8
Error reading segment 9
Index object count mismatch. 614319 != 613381
attic: Exiting with failure status due to previous errors

[whiteadam]

I've figured out with mine that it pretty much happens daily. I have 13 servers backing up over SSH to a central server, with a disk on iSCSI.

I'm not sure how to prevent/remedy these errors on a daily basis yet though.

[tgharold]

This still happens in 0.16. Running "attic check" against the repository results in different segment numbers being reported.

MsgPack version on the origin machine (the one being backed up):

Installed Packages
Name        : python-msgpack
Arch        : x86_64
Version     : 0.4.6
Release     : 1.el6
Size        : 243 k
Repo        : installed
From repo   : epel
Summary     : A Python MessagePack (de)serializer
URL         : http://pypi.python.org/pypi/msgpack-python/
License     : ASL 2.0
Description : MessagePack is a binary-based efficient data interchange format that is
            : focused on high performance. It is like JSON, but very fast and small.
            : This is a Python (de)serializer for MessagePack.

And on the destination machine (housing the repository):

Installed Packages
Name        : python-msgpack
Arch        : x86_64
Version     : 0.4.6
Release     : 1.el6
Size        : 243 k
Repo        : installed
From repo   : epel
Summary     : A Python MessagePack (de)serializer
URL         : http://pypi.python.org/pypi/msgpack-python/
License     : ASL 2.0
Description : MessagePack is a binary-based efficient data interchange format that is
            : focused on high performance. It is like JSON, but very fast and small.
            : This is a Python (de)serializer for MessagePack.

Now, it's possible that corruption crept in before the upgrade to 0.16. I'm running a --repair now.