Syncdaemon silently crashes at startup

[Natalia Bidart]

This is a split from bug #436612.

* Every first time I start syncdaemon (in each boot), the metadata load doesn't finish and SD apparently crashes without leaving any trace. This means that the log file has nothing but:

2010-11-11 08:54:50,330 - ubuntuone.SyncDaemon.VM.MD - DEBUG - metadata version: 6
2010-11-11 08:54:50,385 - ubuntuone.SyncDaemon.fsm - INFO - loading updated metadata

and syncdaemon is not working. So I run u1sdtool -s again, and the log file is rotates and the load metadata process restart, and I get something like this:

2010-11-11 09:04:12,049 - ubuntuone.SyncDaemon.VM.MD - DEBUG - metadata version: 6
2010-11-11 09:04:12,050 - ubuntuone.SyncDaemon.fsm - INFO - loading updated metadata
2010-11-11 09:06:05,100 - ubuntuone.SyncDaemon.fsm - INFO - initialized: idx_path: 19704, idx_node_id: 15127, shares: 12

So the second metadata load process finish because the first attempt cached most of the files in the disk cache. So the second attempt takes 2 minutes when most of the files are already in the disk cache! So this amount of time is extremely large.

[Natalia Bidart (nataliabidart)]

Starting syncdaemon with:

/usr/lib/ubuntuone-client/ubuntuone-syncdaemon --logging-level=DEBUG

will complete the metadata load in the first attempt. In this case, it took:

2010-11-11 13:07:56,933 - ubuntuone.SyncDaemon.VM.MD - DEBUG - metadata version: 6
2010-11-11 13:07:57,021 - ubuntuone.SyncDaemon.fsm - INFO - loading updated metadata
2010-11-11 13:12:45,864 - ubuntuone.SyncDaemon.fsm - INFO - initialized: idx_path: 19704, idx_node_id: 16948, shares: 12

Almost 5 minutes of intensive disk usage (which makes the rest of the desktop pretty unusable due to IO starvation).

[Natalia Bidart (nataliabidart)]

Killing happen after exactly 2 minutes:

nessita@miro:~$ date; u1sdtool --start
Tue Nov 30 11:55:29 ART 2010

nessita@miro:~$ while [ $(ps aux | grep syncdaemon | grep -v grep | wc -l) -gt 0 ]; do date; sleep 1;
done
...
Tue Nov 30 11:57:30 ART 2010

[Guillermo Gonzalez (verterok)]

Hi!

Naty, Thanks for all the testing.

This problem is because syncdaemon is registering dbus with a different name than the specified in the .service file, and dbus-daemon is killing the child process (launched by dbus-activation) because of this.

[Roman Yepishev (rye)]

This happens because I introduced a .UniqueInstance bus name in LP:639017