Hello,
We're running an instance of Phabricator for swiss universities and have some problem with slow imports for large repository (1000+ commits). Here's for instance what happens, the queue has a lot of tasks but the queue usage is very low, usually around 20-30%.
{F1685850}
{F1685852}
Here's the result of phd log, http://pastebin.com/9BN7ncSY
The only thing I see is some errors because some people didn't correctly set up import from github. On our test instance, we don't have those error but still have the import issue.
It happens on both our production instances and test, which are running the following version,
prod
phabricator 9da15fd7ab7071e3b7361463229abd4e918d9d94
libphutil 5fd2cf9d5ddd38424a54a8fba02398d527639970
arcanist f1c45a3323ae20eefe29c0a22c7923fe8b151bbf
test
phabricator be84c051e5ba05296be4070bdfbf6f1fdd44355b
libphutil d6818e59c1764ede22cad56f8c5b1b527cb6a577
arcanist 10e5194752901959507223c01e0878e6b8312cc5
We did our install before the cluster configuration was a thing, so we did our own infra which consist of,
- app instances without daemons
- daemons on dedicated instance (one instance for now)
- mysql galera using only one node as master and other nodes as backup (via haproxy)
- repo files shared accross app instances using glusterfs fuse client
So far we have no other problem, performances are very good.
Thanks a lot for any idea,
Regards
EDIT: An other example, I just pushed a new repo of 620 commits, with 2 phd instances
{F1797336}