On 2025-03-29 at 14:07:17, Akash S wrote: > We compared the GIT_TRACE_PACKET=1 output of successful and > unsuccessful pushes and noticed that the "git flush message 0000" is > not being received from the Azure DevOps Server during unsuccessful > push. > > ------------------------------------------------------------------------------------------------------------------------------------------------------- > > Successful git push: > 11:29:09.032050 pkt-line.c:85 packet: git< ok refs/tags/v3.9.1 > 11:29:09.034601 http.c:878 == Info: TLSv1.3 (IN), TLS app data, [no content] (0): > 11:29:09.034630 pkt-line.c:85 packet: sideband< \10000 > 11:29:09.034643 http.c:878 == Info: TLSv1.3 (IN), TLS app data, [no content] (0): > 11:29:09.034664 pkt-line.c:85 packet: sideband< 0000 > 11:29:09.034664 pkt-line.c:85 packet: git< 0000 > 11:29:09.034699 pkt-line.c:85 packet: git> 0000 > 11:29:09.035667 http.c:878 == Info: TLSv1.3 (IN), TLS app data, [no content] (0): > 11:29:09.035701 http.c:878 == Info: Connection #0 to host dev.azure.com left intact > To https://dev.azure.com/ORGNAME/PRJNAME/_git/REPONAME > > ------------------------------------------------------------------------------------------------------------------------------------------------------- > > Unsuccessful git push: > 11:33:58.759005 pkt-line.c:85 packet: sideband< \10018ok refs/tags/v3.1.7 > 11:33:58.759035 pkt-line.c:85 packet: git< ok refs/tags/v3.1.7 > 11:33:58.760165 http.c:878 == Info: TLSv1.3 (IN), TLS app data, [no content] (0): > 11:33:58.760197 http.c:878 == Info: Connection #0 to host dev.azure.com left intact > (process hung here) > > ------------------------------------------------------------------------------------------------------------------------------------------------------- > > We noticed that "git push mirror" ultimately spawns "git send-pack" > with options "--stdin" and "--stateless-rpc" and from the git > send-pack documentation we confirmed that the line must end with a > flush packet which is ultimately not happening in the hung cases. > pstack of send-pack process also shows it is waiting on reading > packets as you mentioned earlier. `--stateless-rpc` is used for HTTPS, so that's expected. > Can you please confirm if the hung issue is due to the missing message or packet from the Azure DevOps Server? We don't have a full trace output here, so it's hard to say completely for certain, but from what I know of the protocol off the top of my head, a flush packet is always required at the end of the section of message. So I would say that it's very likely that's the cause here. Note that it could be that you're _also_ missing one or more `ok` responses from the ref updates you're doing here, and that it's hanging _before_ it would normally send the flush packet. That might point to a different cause on the server side, but the effect would be the same. If you're talking to the Azure DevOps people, I'd provide them a full trace output of both working and non-working pushes so that they can inspect it themselves, which will help them figure out where it might be going wrong. -- brian m. carlson (they/them) Toronto, Ontario, CA
Attachment:
signature.asc
Description: PGP signature
