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

Re: fetching packs and storing them as packs

From
Junio C Hamano <junkio@cox.net>
Date
Oct 28, 2006, 05:42 UTC
Message-ID
<7vwt6l9etn.fsf@assigned-by-dhcp.cox.net>
In-Reply-To
<Pine.LNX.4.64.0610272109500.3849@g5.osdl.org>
Linus Torvalds <torvalds@osdl.org> writes:
Show 20 quoted lines
> On Fri, 27 Oct 2006, Shawn Pearce wrote:
>> 
>> So a reader-writer lock is preferred over
>> a non-locking solution such as I posted in
>> http://article.gmane.org/gmane.comp.version-control.git/30288 ?
>> 
>> Not to mention that such a solution would also fix the -d issue
>> Linus points out above.
>
> Be very careful.
>
> There's a good reason why git doesn't use locking, and tends to use the 
> "create file exclusively and move over the old version after having tested 
> that the old version is still relevant" approach.
>
> Two _major_ issues:
>...
>
> I would MUCH rather we just rename the index/pack file to something that 
> git can _use_, but that "git repack -a -d" won't remove....
Two points.

The "locking" I mentioned was between receive-pack and repack -a -d; upload-pack (what millions people are using to read from the repository you are pushing into) is not affected. So in that sense, we can afford to use lock without much contention.

I just thought of a cute hack that does not involve renaming packs at all (so no need to match new-pack-X.pack with pack-X.idx), and Shawn's sequence actually would work, which is:

The receive-pack side:
  a. Create temporary pack file in $GIT_DIR/objects/pack_XXXXX.
  b. Create temporary index file in $GIT_DIR/objects/index_XXXXX.
  c. Write pack and index, in "inactive" state.
  d. Move pack to $GIT_DIR/objects/pack/...
  e. Move idx to $GIT_DIR/objects/pack...
  f. Update refs.
  g. Mark new pack and idx as "active".
The "repack -a -d" side:
  1. List all active packs and store in memory.
  2. Repack only loose objects and objects contained in active packs.
  3. Move new pack and idx into $GIT_DIR/objects/pack/...
  4. Mark new pack and idx as "active".
  5. Delete active packs found by step #1.

Pack-idx pair is marked "active" by "chmod u+s" the .pack file. During the normal operation, all .pack/.idx pair in objects/pack/ directories are usable regardless of the setuid bit; we would never make .pack files executable so u+s would not otherwise hurt us either. "active" probably is better read as "eligible for repacking".

Previous: Linus TorvaldsNext: Shawn Pearce
Message 8 of 51 in “fetching packs and storing them as packs”
  1. Nicolas PitreOct 26, 2006
  2. Eran TromerOct 26, 2006
  3. Linus TorvaldsOct 27, 2006
  4. Junio C HamanoOct 27, 2006
  5. Shawn PearceOct 28, 2006
  6. Junio C HamanoOct 28, 2006
  7. Linus TorvaldsOct 28, 2006
  8. Junio C HamanoOct 28, 2006
  9. Shawn PearceOct 28, 2006
  10. Shawn PearceOct 28, 2006
  11. Junio C HamanoOct 28, 2006
  12. Shawn PearceOct 29, 2006
  13. Junio C HamanoOct 29, 2006
  14. Shawn PearceOct 29, 2006
  15. Junio C HamanoOct 29, 2006
  16. Shawn PearceOct 29, 2006
  17. Linus TorvaldsOct 28, 2006
  18. Junio C HamanoOct 28, 2006
  19. Eran TromerOct 28, 2006
  20. Shawn PearceOct 29, 2006
  21. Jakub NarebskiOct 29, 2006
  22. Shawn PearceOct 29, 2006
  23. send-pack --keep: do not explode into loose objects on the receiving end.Junio C Hamano, Oct 29, 2006
  24. Shawn PearceOct 29, 2006
  25. Junio C HamanoOct 29, 2006
  26. Nicolas PitreOct 30, 2006
  27. Eran TromerOct 26, 2006
  28. Nicolas PitreOct 27, 2006
  29. Shawn PearceOct 27, 2006
  30. SeanOct 27, 2006
  31. Junio C HamanoOct 27, 2006
  32. Nicolas PitreOct 27, 2006
  33. Nicolas PitreOct 27, 2006
  34. Eran TromerOct 27, 2006
  35. Shawn PearceOct 27, 2006
  36. SeanOct 27, 2006
  37. Jakub NarebskiOct 27, 2006
  38. SeanOct 27, 2006
  39. Eran TromerOct 27, 2006
  40. Shawn PearceOct 27, 2006
  41. Alex RiesenOct 27, 2006
  42. Shawn PearceOct 27, 2006
  43. Alex RiesenOct 27, 2006
  44. Shawn PearceOct 27, 2006
  45. Nicolas PitreOct 27, 2006
  46. Petr BaudisOct 27, 2006
  47. J. Bruce FieldsOct 27, 2006
  48. Petr BaudisOct 27, 2006
  49. J. Bruce FieldsOct 27, 2006
  50. J. Bruce FieldsOct 27, 2006
  51. Junio C HamanoOct 27, 2006

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.