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

[PATCH v2 00/10] upload-pack: reduce lock contention when writing packfile data

From
Patrick Steinhardt <ps@pks.im>
Date
Mar 3, 2026, 15:00 UTC
Message-ID
<20260303-pks-upload-pack-write-contention-v2-0-7321830f08fe@pks.im>
In-Reply-To
<20260227-pks-upload-pack-write-contention-v1-0-7166fe255704@pks.im>
Hi,

this small patch series fixes some heavy lock contention when writing data from git-upload-pack(1) into pipes. This lock contention can be observed when having hundreds of git-upload-pack(1) processes active at the same time that write data into pipes at dozens of gigabits per second.

I have uploaded the flame graph that clearly shows the lock contention at [1].

Changes in v2:
  - Change the buffer size in git-pack-objects(1) to also reduce the
    number of write syscalls over there.
  - Introduce writev to half the number of syscalls when writing
    pktlines.
  - Use `sizeof(os->buffer)` instead of open-coding its size.
  - Improve keepalive logic in git-upload-pack(1) to account for
    buffering.
  - Link to v1: https://lore.kernel.org/r/20260227-pks-upload-pack-write-contention-v1-0-7166fe255704@pks.im
Thanks!
Patrick
[1]: https://gitlab.com/gitlab-org/git/-/work_items/675
---
Patrick Steinhardt (10):
      upload-pack: fix debug statement when flushing packfile data
      upload-pack: adapt keepalives based on buffering
      upload-pack: reduce lock contention when writing packfile data
      git-compat-util: introduce `cast_size_t_to_ssize_t()`
      compat/posix: introduce writev(3p) wrapper
      wrapper: introduce writev(3p) wrappers
      sideband: use writev(3p) to send pktlines
      csum-file: introduce `hashfd_ext()`
      csum-file: drop `hashfd_throughput()`
      builtin/pack-objects: reduce lock contention when writing packfile data
 Makefile               |  4 ++++
 builtin/pack-objects.c | 23 +++++++++++++++----
 compat/posix.h         | 14 ++++++++++++
 compat/writev.c        | 29 ++++++++++++++++++++++++
 config.mak.uname       |  2 ++
 csum-file.c            | 28 +++++++----------------
 csum-file.h            | 16 ++++++++++++--
 git-compat-util.h      |  8 +++++++
 meson.build            |  1 +
 sideband.c             | 14 +++++++++---
 upload-pack.c          | 60 ++++++++++++++++++++++++++++++++++++++++----------
 wrapper.c              | 41 ++++++++++++++++++++++++++++++++++
 wrapper.h              |  9 ++++++++
 write-or-die.c         |  8 +++++++
 write-or-die.h         |  1 +
 15 files changed, 217 insertions(+), 41 deletions(-)
Range-diff versus v1:
 1:  368a984880 =  1:  81c00e3b52 upload-pack: fix debug statement when flushing packfile data
 -:  ---------- >  2:  01c86a9573 upload-pack: adapt keepalives based on buffering
 2:  f014e8005a !  3:  16bc5661d4 upload-pack: reduce lock contention when writing packfile data
    @@ Commit message
         write(3p) syscalls from 130,000 to 100,000. So overall this doesn't seem
         worth it.
     
    +    Note that the issue could also be fixed by adapting the write buffer
    +    that we use in the downstream git-pack-objects(1) command, and such a
    +    change would have roughly the same result. But the command that
    +    generates the packfile data may not always be git-pack-objects(1) as it
    +    can be changed via "uploadpack.packObjectsHook", so such a fix would
    +    only help in _some_ cases. Regardless of that, we'll also adapt the
    +    write buffer size of git-pack-objects(1) in a subsequent commit.
    +
         Helped-by: Matt Smiley <msmiley@gitlab.com>
         Signed-off-by: Patrick Steinhardt <ps@pks.im>
     
    @@ upload-pack.c: static int relay_pack_data(int pack_objects_out, struct output_st
     +	 * Make sure that we buffer some data before sending it to the client.
     +	 * This significantly reduces the number of write(3p) syscalls.
     +	 */
    -+	if (readsz && os->used < (LARGE_PACKET_DATA_MAX * 2 / 3))
    ++	if (readsz && os->used < (sizeof(os->buffer) * 2 / 3))
     +		return readsz;
     +
      	if (os->used > 1) {
 -:  ---------- >  4:  f42a8a0558 git-compat-util: introduce `cast_size_t_to_ssize_t()`
 -:  ---------- >  5:  830c72f588 compat/posix: introduce writev(3p) wrapper
 -:  ---------- >  6:  733ef129f1 wrapper: introduce writev(3p) wrappers
 -:  ---------- >  7:  18f7429e14 sideband: use writev(3p) to send pktlines
 -:  ---------- >  8:  8eb2aee74d csum-file: introduce `hashfd_ext()`
 -:  ---------- >  9:  a634506b81 csum-file: drop `hashfd_throughput()`
 -:  ---------- > 10:  ca6d52b012 builtin/pack-objects: reduce lock contention when writing packfile data

--- base-commit: 7b2bccb0d58d4f24705bf985de1f4612e4cf06e5 change-id: 20260227-pks-upload-pack-write-contention-435ce01f5fe9

Previous: Patrick SteinhardtNext: Patrick Steinhardt
Message 13 of 53 in “upload-pack: reduce lock contention when writing packfile data”
  1. 0/2 upload-pack: reduce lock contention when writing packfile dataPatrick Steinhardt, Feb 27, 2026
  2. 1/2 upload-pack: fix debug statement when flushing packfile dataPatrick Steinhardt, Feb 27, 2026
  3. 2/2 upload-pack: reduce lock contention when writing packfile dataPatrick Steinhardt, Feb 27, 2026
  4. brian m. carlsonFeb 27, 2026
  5. Patrick SteinhardtFeb 27, 2026
  6. Junio C HamanoFeb 27, 2026
  7. Jeff KingFeb 27, 2026
  8. Patrick SteinhardtMar 2, 2026
  9. Jeff KingMar 2, 2026
  10. Patrick SteinhardtMar 3, 2026
  11. Jeff KingMar 3, 2026
  12. Patrick SteinhardtMar 3, 2026
  13. 00/10 upload-pack: reduce lock contention when writing packfile dataPatrick Steinhardt, Mar 3, 2026
  14. 01/10 upload-pack: fix debug statement when flushing packfile dataPatrick Steinhardt, Mar 3, 2026
  15. 02/10 upload-pack: adapt keepalives based on bufferingPatrick Steinhardt, Mar 3, 2026
  16. Jeff KingMar 5, 2026
  17. Patrick SteinhardtMar 10, 2026
  18. 03/10 upload-pack: reduce lock contention when writing packfile dataPatrick Steinhardt, Mar 3, 2026
  19. Jeff KingMar 5, 2026
  20. Patrick SteinhardtMar 10, 2026
  21. 04/10 git-compat-util: introduce `cast_size_t_to_ssize_t()`Patrick Steinhardt, Mar 3, 2026
  22. 05/10 compat/posix: introduce writev(3p) wrapperPatrick Steinhardt, Mar 3, 2026
  23. Junio C HamanoMar 4, 2026
  24. Jeff KingMar 5, 2026
  25. brian m. carlsonMar 5, 2026
  26. Johannes SixtMar 5, 2026
  27. brian m. carlsonMar 5, 2026
  28. Patrick SteinhardtMar 10, 2026
  29. 06/10 wrapper: introduce writev(3p) wrappersPatrick Steinhardt, Mar 3, 2026
  30. 07/10 sideband: use writev(3p) to send pktlinesPatrick Steinhardt, Mar 3, 2026
  31. Junio C HamanoMar 4, 2026
  32. 08/10 csum-file: introduce `hashfd_ext()`Patrick Steinhardt, Mar 3, 2026
  33. Junio C HamanoMar 4, 2026
  34. Patrick SteinhardtMar 10, 2026
  35. 09/10 csum-file: drop `hashfd_throughput()`Patrick Steinhardt, Mar 3, 2026
  36. 10/10 builtin/pack-objects: reduce lock contention when writing packfile dataPatrick Steinhardt, Mar 3, 2026
  37. 00/10 upload-pack: reduce lock contention when writing packfile dataPatrick Steinhardt, Mar 10, 2026
  38. 01/10 upload-pack: fix debug statement when flushing packfile dataPatrick Steinhardt, Mar 10, 2026
  39. 02/10 upload-pack: adapt keepalives based on bufferingPatrick Steinhardt, Mar 10, 2026
  40. 03/10 upload-pack: prefer flushing data over sending keepalivePatrick Steinhardt, Mar 10, 2026
  41. Junio C HamanoMar 10, 2026
  42. Patrick SteinhardtMar 10, 2026
  43. 04/10 upload-pack: reduce lock contention when writing packfile dataPatrick Steinhardt, Mar 10, 2026
  44. 05/10 compat/posix: introduce writev(3p) wrapperPatrick Steinhardt, Mar 10, 2026
  45. Junio C HamanoMar 10, 2026
  46. 06/10 wrapper: introduce writev(3p) wrappersPatrick Steinhardt, Mar 10, 2026
  47. 07/10 sideband: use writev(3p) to send pktlinesPatrick Steinhardt, Mar 10, 2026
  48. 08/10 csum-file: introduce `hashfd_ext()`Patrick Steinhardt, Mar 10, 2026
  49. 09/10 csum-file: drop `hashfd_throughput()`Patrick Steinhardt, Mar 10, 2026
  50. 10/10 builtin/pack-objects: reduce lock contention when writing packfile dataPatrick Steinhardt, Mar 10, 2026
  51. Junio C HamanoMar 10, 2026
  52. Johannes SixtMar 10, 2026
  53. Patrick SteinhardtMar 11, 2026

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.