Re: [PATCH datacenter-manager 08/15] task cache: archive iterator: don't yield more items if reading from file failed

["Lukas Wagner" <l.wagner@proxmox.com>]

On Fri Jul 10, 2026 at 1:36 PM CEST, Dominik Csapak wrote:
> sorry maybe i'm not deep enough into the code here but
> how would a single err result produce an endless loop?
>
> the underlying lines iterator should not infinitely return the error?
>
> maybe a bit more context for that issue in the commit message would
> be nice.... (or maybe it's obvious enough for other, just not me ;) )

Fair enough, I've rewritten the commit message to include a more detail:

  task cache: archive iterator: don't yield more items if reading from file failed

  When iterating over lines in a file with Lines<Box<dyn BufRead>> and the
  BufRead is based on a zstd Decoder, a malformed zstd file will produce
  an endless stream of Errors during iteration; apparently the read
  position inside the decoder is never advanced and thus never reaches
  the end of the file.

  To avoid running into an endless loop, don't yield any more items from
  ArchiveIterator after the first read has failed.

  Also add a test case that verifies the absence of the issue.

>
> also should we handle partial corruption, e.g. when only a single line
> can't be read? (this skips the rest IIUC)

I think it's fine to skip the rest of the file if the read_line fails,
since that is probably caused by an IO error or by a malformed zstd
file - so we cannot really read anything anyways.

If only a single line could not be decoded (e.g. malformed JSON),
the later introduced recovery code will skip it and mark the file for
recovery.