 | Paul Sokolovsky Oct 15 23:45 On Mon, 13 Oct 2014 11:03:09 +0000 "Benjamin Copeland (Linaro IT Services)" <its@linaro.org> wrote: Hello Christoffer, Mark, Jon, Sorry, these are weird issues, which is hard to debug post-factum. As the matter was about /git-ro/ (dump git protocol), the likely cause is indeed "git update-server-info" haven't been run. Running it is however handled by gitolite, via post-update hook. So, how could it happen it didn't run? Well, unfortunately we saw such cases in the past. We tried to debug them (e.g. I remember adding logging to update hook), but as it's pretty lurky thing, we didn't had much success with it. Suffice to say that under some circumstances (heavy load) process creation in Linux can fail, and knowing that git servers are under regular load, that seems to be it. Instead, for android.git.linaro.org, we went to add 2nd layer of reliability - run "git update-server-info" via cron too. I checked and see that git.linaro.org lacks this cronjob, so let me know if you'd feel more comfortable with it. But as Ben already mentioned, the suggested solution is to use smart HTTP protocol (URLs w/o /git-ro/). /git-ro/ is still listed in the gitweb just to offer people a choice in case of issues or peculiar circumstances, but otherwise on a gitweb page like https://git.linaro.org/people/christoffer.dall/linux-kvm-arm.git clone URLs are listed in the order of suggested use preference. Hope that clarifies the situation, and feel free to ping me (pfalcon) directly on IRC if you spot anything like that again, so I can see what's up while it's hot. Thanks! |