git/list[1] front-page[2] threads[3] people[4] search[5] about
 

Re: [PATCH] packed_ref_store: handle a packed-refs file that is a symlink

From
Junio C Hamano <gitster@pobox.com>
Date
Jul 27, 2017, 19:40 UTC
Message-ID
<xmqqlgn9mzww.fsf@gitster.mtv.corp.google.com>
In-Reply-To
<20170727182854.wxjvjs2x746n3x2t@sigill.intra.peff.net>
Jeff King <peff@peff.net> writes:
Show 10 quoted lines
> On Thu, Jul 27, 2017 at 10:19:48AM -0700, Junio C Hamano wrote:
>
>> Makes sense.  Makes me wonder why we need a separate .new file
>> (instead of writing into the .lock instead), but that is a different
>> issue.
>
> It comes from 42dfa7ece (commit_packed_refs(): use a staging file
> separate from the lockfile, 2017-06-23). That commit explains that we
> want to be able to put the new contents into service before we release
> the lock. But it doesn't say why that's useful.

By being able to hold the lock on packed-refs longer, I guess something like this becomes possible:

 * hold the lock on packed-refs
 * hold the lock on loose ref A, B, C, ...
 * update packed-refs to include the freshest values of these refs
 * start serving packed-refs without releasing the lock
 * for X in A, B, C...: delete the loose ref X and unlock X
 * unlock the packed-refs

Other people racing with the sequence to recreate a loose ref that is even fresher than the resulting packed-refs file, while we still hold the lock on packed-refs, is perfectly OK.

But we must make sure our packed-refs is visible to others before starting to delete and unlock the loose refs.

Hmph, but that is not a sufficient explanation. I am not seeing anything bad to happen if we unlock the packed-refs before deleting loose refs that we have locks on, so there must be something else that needs "new packed-refs is made visible way before we unlock it".

Show 7 quoted lines
> I recall from past discussions that this will help close some races,
> and e5cc7d7d2 (repack_without_refs(): don't lock or unlock the packed
> refs, 2017-07-01) alludes to this. I think the races in question have to
> do with holding the packed-refs lock while pruning the just-packed
> files, but I'm having trouble digging up specifics in the archive.
>
> -Peff
Previous: Jeff KingNext: Michael Haggerty
Message 41 of 52 in “Create a reference backend for packed refs”
  1. 00/30 Create a reference backend for packed refsMichael Haggerty, Jul 1, 2017
  2. 01/30 t1408: add a test of stale packed refs covered by loose refsMichael Haggerty, Jul 1, 2017
  3. 03/30 packed_ref_store: new structMichael Haggerty, Jul 1, 2017
  4. 04/30 packed_ref_store: move `packed_refs_path` hereMichael Haggerty, Jul 1, 2017
  5. 02/30 add_packed_ref(): teach function to overwrite existing refsMichael Haggerty, Jul 1, 2017
  6. 06/30 clear_packed_ref_cache(): take a `packed_ref_store *` parameterMichael Haggerty, Jul 1, 2017
  7. 05/30 packed_ref_store: move `packed_refs_lock` member hereMichael Haggerty, Jul 1, 2017
  8. 09/30 get_packed_refs(): take a `packed_ref_store *` parameterMichael Haggerty, Jul 1, 2017
  9. 07/30 validate_packed_ref_cache(): take a `packed_ref_store *` parameterMichael Haggerty, Jul 1, 2017
  10. 12/30 commit_packed_refs(): take a `packed_ref_store *` parameterMichael Haggerty, Jul 1, 2017
  11. 08/30 get_packed_ref_cache(): take a `packed_ref_store *` parameterMichael Haggerty, Jul 1, 2017
  12. 10/30 add_packed_ref(): take a `packed_ref_store *` parameterMichael Haggerty, Jul 1, 2017
  13. 11/30 lock_packed_refs(): take a `packed_ref_store *` parameterMichael Haggerty, Jul 1, 2017
  14. 16/30 packed_peel_ref(): new function, extracted from `files_peel_ref()`Michael Haggerty, Jul 1, 2017
  15. 15/30 repack_without_refs(): take a `packed_ref_store *` parameterMichael Haggerty, Jul 1, 2017
  16. 14/30 get_packed_ref(): take a `packed_ref_store *` parameterMichael Haggerty, Jul 1, 2017
  17. 17/30 packed_ref_store: support iterationMichael Haggerty, Jul 1, 2017
  18. 19/30 packed-backend: new module for handling packed referencesMichael Haggerty, Jul 1, 2017
  19. 18/30 packed_read_raw_ref(): new function, replacing `resolve_packed_ref()`Michael Haggerty, Jul 1, 2017
  20. 22/30 commit_packed_refs(): use a staging file separate from the lockfileMichael Haggerty, Jul 1, 2017
  21. 20/30 packed_ref_store: make class into a subclass of `ref_store`Michael Haggerty, Jul 1, 2017
  22. 13/30 rollback_packed_refs(): take a `packed_ref_store *` parameterMichael Haggerty, Jul 1, 2017
  23. 23/30 packed_refs_lock(): function renamed from lock_packed_refs()Michael Haggerty, Jul 1, 2017
  24. 28/30 repack_without_refs(): don't lock or unlock the packed refsMichael Haggerty, Jul 1, 2017
  25. 29/30 t3210: add some tests of bogus packed-refs file contentsMichael Haggerty, Jul 1, 2017
  26. 26/30 clear_packed_ref_cache(): don't protest if the lock is heldMichael Haggerty, Jul 1, 2017
  27. 30/30 read_packed_refs(): die if `packed-refs` contains bogus dataMichael Haggerty, Jul 1, 2017
  28. 27/30 commit_packed_refs(): remove call to `packed_refs_unlock()`Michael Haggerty, Jul 1, 2017
  29. 21/30 commit_packed_refs(): report errors rather than dyingMichael Haggerty, Jul 1, 2017
  30. 25/30 packed_refs_unlock(), packed_refs_is_locked(): new functionsMichael Haggerty, Jul 1, 2017
  31. 24/30 packed_refs_lock(): report errors via a `struct strbuf *err`Michael Haggerty, Jul 1, 2017
  32. Jeff KingJul 5, 2017
  33. Stefan BellerJul 20, 2017
  34. Jonathan NiederJul 20, 2017
  35. packed_ref_store: handle a packed-refs file that is a symlinkMichael Haggerty, Jul 26, 2017
  36. Stefan BellerJul 27, 2017
  37. Jonathan NiederJul 27, 2017
  38. Michael HaggertyJul 27, 2017
  39. Junio C HamanoJul 27, 2017
  40. Jeff KingJul 27, 2017
  41. Junio C HamanoJul 27, 2017
  42. Michael HaggertyJul 28, 2017
  43. Ævar Arnfjörð BjarmasonMay 31, 2021
  44. Jeff KingJun 3, 2021
  45. t: use portable wrapper for readlink(1)Jeff King, Jun 3, 2021
  46. Ævar Arnfjörð BjarmasonJun 4, 2021
  47. Felipe ContrerasJun 3, 2021
  48. Jeff KingJun 3, 2021
  49. Felipe ContrerasJun 3, 2021
  50. Ævar Arnfjörð BjarmasonJun 4, 2021
  51. Felipe ContrerasJun 5, 2021
  52. Ævar Arnfjörð BjarmasonJun 4, 2021

Read the whole thread, see it on lore, or plain text.

$ cat FOOTERMessages come from the public archive at lore.kernel.org/git, fetched every hour. The front page is chosen and written each morning by an AI editor and can be wrong; the threads themselves are the record. About and API. For agents: an MCP server at https://gitlist.dev/mcp, and any thread, story or person page as Markdown by adding .md to its URL (or sending Accept: text/markdown). Details in /llms.txt.