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%.
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,
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,
EDIT: An other example, I just pushed a new repo of 620 commits, with 2 phd instances