GNOME Bugzilla – Bug 617170
Other Programs Can Encounter "Disk Full" Problems Whilst Tracker is Mining Data
Last modified: 2010-09-05 20:22:47 UTC
Fedora bug https://bugzilla.redhat.com/show_bug.cgi?id=587107 Description of problem: I tried to manually launch abrt whilst tracker mining/indexing was switched on. abrt would fail with "disk full" errors, though df reported my disk was not full. As soon as tracker was switched off, the problem disappeared. Version-Release number of selected component (if applicable): tracker 0.8.3-1 abrt 1.0.9-2 How reproducible: Only tried once so far. Steps to Reproduce: 1. tracker-control -s (only default miners are installed. Left it running over night to index system. It appeared to have done so as it had reached 100% indexing completed.) 2. service abrtd stop 3. abrtd -dvvv 4. Output of abrt: abrtd: inotify_add_watch failed on '/var/cache/abrt': No space left on device 5. Because of this error, I ran df. Here is the output: Filesystem 1K-blocks Used Available Use% Mounted on /dev/mapper/VolGroup-lv_root 100791728 12018816 83652912 13% / tmpfs 2001184 956 2000228 1% /dev/shm /dev/sda1 198337 98969 89128 53% /boot /dev/mapper/VolGroup-LogVol02 47807384 36117772 9261096 80% /home 6. Next, I turned off the data-mining with: tracker-control -t 7. Ran abrtd -dvvv again. 8. Error no longer appears. Actual results: abrt was reporting that disk was out of space when only 13% of root was in use. Expected results: Data mining/indexing should not consume this amount of resources (if it was) or interfere with running of other programs.
(In reply to comment #0) > abrtd: inotify_add_watch failed on '/var/cache/abrt': No space left on device "No space left on device" is slightly inaccurate. What it really means is, you have no more inotify watches available. You have exhausted them all. Tracker uses MAX - 512 so you always have 512 for other applications, so presumably some other application is using a lot of inotify watches AND you used up all the watches in Tracker too. > 5. Because of this error, I ran df. Here is the output: > > Filesystem 1K-blocks Used Available Use% Mounted on > /dev/mapper/VolGroup-lv_root > 100791728 12018816 83652912 13% / > tmpfs 2001184 956 2000228 1% /dev/shm > /dev/sda1 198337 98969 89128 53% /boot > /dev/mapper/VolGroup-LogVol02 > 47807384 36117772 9261096 80% /home Yea, this has nothing to do with actual disk space. > 6. Next, I turned off the data-mining with: tracker-control -t > 7. Ran abrtd -dvvv again. > 8. Error no longer appears. That's because tracker uses a lot if watches (depending on your file system). Are you able to provide logs from tracker-miner-fs? For details see the Logging section of this page: http://live.gnome.org/Tracker/Documentation/Debugging That log should tell you if you're exhausting your inotify watches or not, which I suspect you are, in any case, there should be *some* left for your application to operate normally unless another application uses too many too.
Martyn, is there any way to know which other applications are taking the inotify watches? Probably abrt only needs a single watch in /var/cache/abrt to get notified when new coredumps arrive to the folder.
(In reply to comment #2) > Martyn, is there any way to know which other applications are taking the > inotify watches? Probably abrt only needs a single watch in /var/cache/abrt to > get notified when new coredumps arrive to the folder. Nope, not that I know of.
Closing this bug report as no further information has been provided. Please feel free to reopen this bug if you can provide the information asked for. Thanks!