[Borgbackup] Borg filling local disk completely during backup

[Thomas Waldmann]

On 2020-05-16 17:13, Simon Hoffmann via Borgbackup wrote:
>>> Looking at the drive, I found ~50 GB files in /root/.cache/borg/xxx/chunks.archive.d. (yes, borg runs as root)
>>
>> This is documented, search the docs for chunks.archive.dir and the misc.
>> BORG_*_DIR settings.
> 
> Yeah, but I'm still not really sure why this happened. For weeks, the disk usage did
> not change at all (and thus chunks.archive.d dig not get any bigger), and all of a
> sudden borg uses that directory and fills up the entire disk within minutes.

If borg detects that the repo is newer than its local chunks index, it
must rebuild this index.

To do that, it downloads metadata of all archives and builds per-archive
indexes and caches them to chunks.archive.d/ (so that, if that happens
again, it only needs to download the new archives' metadata).

>From all these smaller indexes, it builds its master chunks index.

One common cause that triggers this is accessing the repo from some
other client, some other user or locally on the repo server.

> The only thing I could find about chunks.archive.d in the manual was in
> internals/data-structure. So the reason for this behaviour is that all of a sudden
> the whole repo became corrupted and borg had to rebuild all indices?

It's about cache coherency, not about corruption.

Of course, if your local index gets corrupt somehow and you need to kill
it, you will also have to rebuild it.

-- 

GPG ID: 9F88FB52FAF7B393
GPG FP: 6D5B EF9A DD20 7580 5747 B70F 9F88 FB52 FAF7 B393