#7029 closed defect (worksforme)
Slow trac performance
| Reported by: | jerk | Owned by: | Jonas Borgström |
|---|---|---|---|
| Priority: | normal | Milestone: | |
| Component: | web frontend | Version: | 0.11b2 |
| Severity: | major | Keywords: | cgi |
| Cc: | trac@… | Branch: | |
| Release Notes: | |||
| API Changes: | |||
| Internal Changes: | |||
Description
just have multiple instances of trac projects running with trac 0.11b2 and do have a very bad performance. the db is about 500MB big, but according to top not the bottleneck. moreover the trac.cgi process creates a 100% load for about 30-45 seconds before the page is displayed and finally the machine is a 512 MB, 1.6 GHz Xeon Processor.
Secondly I tried running tracd alternatively to my mod_python/apache setup and had the same performance and also deactivating the plugins did not significantly improve the sites performance.
Attachments (0)
Change History (17)
follow-up: 3 comment:1 by , 18 years ago
| Type: | enhancement → defect |
|---|
comment:2 by , 18 years ago
| Severity: | normal → major |
|---|
#7035 closed as duplicate. The increased time needed to render big changesets can even lead to timeouts in Safari.
comment:3 by , 18 years ago
Replying to cboos:
Chances are that with only 512MB available the server is swapping, as there are known memory issues (see #6614). Can you move the server to a machine with more RAM? (at least 1GB). If not, you should at least use Python 2.5 which is somewhat better at giving memory back to the system.
The machine is not swapping. I checked this already.
A "100% load for about 30-45 second" is unfortunately not uncommon with lengthy pages like big changesets now that they're rendered with Genshi. There are ongoing efforts to speed things up at the Genshi level, including a GSoC proposal.
In order to be able to do something more specific about your issue, we need to know additional details: for which pages do you see this bad performance, all? some? If it's the timeline, you can reduce the default "daysback" parameter. For big changesets, there are other parameters to tune (TracIni#changeset-section), etc.
The performance issue is esp. regarding viewing and creating tickets. It is in no way related to view change sets.
comment:4 by , 18 years ago
My bad, too much time spent on debugging memory related issues :-)
So OK, your particular problem has probably nothing to do with the rendering. Are you sure this happens also for simply viewing tickets, not only when creating and modifying them? If this is the latter case, then I strongly suspect issues related to the notification (like in #3220). Please try disabling ticket notifications and if this is indeed the problem, check your SMTP server configuration.
comment:5 by , 18 years ago
I just tried this, but it did not help. I disabled the whole notification settings and created a new ticket. this was still very slow. further more things like viewing the report list and viewing some ticket lists is slow.
though I still think this is more like a general issue. I also checked the performance of the timeline, wiki, roadmap, admin etc… and the are all slow…
comment:6 by , 18 years ago
here is just my top output when I'm loading some trac page… does this give you a better idea?
top - 10:37:56 up 9 days, 20:04, 1 user, load average: 0.39, 0.80, 0.56
Tasks: 45 total, 4 running, 41 sleeping, 0 stopped, 0 zombie
Cpu(s): 16.6%us, 10.3%sy, 0.0%ni, 72.8%id, 0.0%wa, 0.0%hi, 0.0%si, 0.3%st
Mem: 524440k total, 459008k used, 65432k free, 99456k buffers
Swap: 0k total, 0k used, 0k free, 195188k cached
Unknown command - try 'h' for help
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
30987 www-data 17 0 20080 11m 3648 R 26.3 2.3 0:00.79 trac.cgi
1 root 16 0 1952 664 568 S 0.0 0.1 0:00.24 init
2 root RT 0 0 0 0 S 0.0 0.0 0:00.00 migration/0
3 root 34 19 0 0 0 R 0.0 0.0 0:00.53 ksoftirqd/0
4 root RT 0 0 0 0 S 0.0 0.0 0:00.00 watchdog/0
5 root 10 -5 0 0 0 S 0.0 0.0 0:00.33 events/0
6 root 10 -5 0 0 0 S 0.0 0.0 0:00.00 khelper
7 root 10 -5 0 0 0 S 0.0 0.0 0:00.00 kthread
8 root 10 -5 0 0 0 S 0.0 0.0 0:00.00 xenwatch
9 root 10 -5 0 0 0 S 0.0 0.0 0:00.00 xenbus
14 root 10 -5 0 0 0 S 0.0 0.0 0:00.00 kblockd/0
44 root 15 0 0 0 0 S 0.0 0.0 0:46.37 pdflush
45 root 15 0 0 0 0 S 0.0 0.0 0:53.27 pdflush
47 root 20 -5 0 0 0 S 0.0 0.0 0:00.00 aio/0
46 root 15 0 0 0 0 S 0.0 0.0 0:00.49 kswapd0
568 root 17 -5 0 0 0 S 0.0 0.0 0:00.00 kseriod
1492 root 15 0 0 0 0 S 0.0 0.0 0:41.50 kjournald
1754 root 12 -5 0 0 0 S 0.0 0.0 0:00.00 kmirrord
1938 root 16 0 1752 560 388 S 0.0 0.1 0:00.13 dhclient
2073 root 16 0 1632 640 508 S 0.0 0.1 0:02.92 syslogd
2080 root 16 0 1584 400 328 S 0.0 0.1 0:00.00 klogd
2104 postgres 16 0 19576 2284 1900 S 0.0 0.4 0:29.48 postmaster
2107 postgres 15 0 10376 1856 484 S 0.0 0.4 0:24.53 postmaster
2108 postgres 15 0 9520 1080 584 S 0.0 0.2 0:17.68 postmaster
2120 postgres 16 0 4524 1720 1392 S 0.0 0.3 0:54.71 pg_autovacuum
2173 Debian-e 16 0 5340 988 692 S 0.0 0.2 0:00.09 exim4
2197 root 16 0 4932 1100 768 S 0.0 0.2 0:00.00 sshd
2229 daemon 15 0 1836 424 300 S 0.0 0.1 0:00.00 atd
2236 root 16 0 2196 896 712 S 0.0 0.2 0:00.94 cron
2301 root 16 0 6440 4744 1556 S 0.0 0.9 0:07.28 munin-node
2330 root 17 0 1576 500 432 S 0.0 0.1 0:00.00 getty
21860 root 16 0 24028 9112 5192 S 0.0 1.7 0:00.64 apache2
24420 www-data 16 0 29744 12m 2860 S 0.0 2.3 0:01.34 apache2
15150 www-data 16 0 30116 11m 2852 S 0.0 2.3 0:01.06 apache2
15457 www-data 17 0 35656 17m 2856 S 0.0 3.4 0:06.60 apache2
15481 www-data 15 0 28128 10m 2868 S 0.0 2.1 0:04.33 apache2
15492 www-data 17 0 32512 14m 2580 S 0.0 2.9 0:05.20 apache2
15493 www-data 16 0 34636 16m 2868 S 0.0 3.2 0:12.34 apache2
15566 www-data 16 0 27696 10m 2864 S 0.0 2.0 0:00.62 apache2
15567 www-data 16 0 32320 14m 2860 S 0.0 2.8 0:04.99 apache2
15576 www-data 16 0 27468 9840 2580 S 0.0 1.9 0:00.66 apache2
2981 www-data 16 0 24164 5644 1500 S 0.0 1.1 0:00.03 apache2
30611 root 16 0 7864 2336 1888 R 0.0 0.4 0:00.05 sshd
30614 root 15 0 5104 1736 1336 S 0.0 0.3 0:00.01 bash
30986 root 16 0 2232 1112 860 R 0.0 0.2 0:00.00 top
comment:7 by , 18 years ago
Well, first don't expect to get any good performance when using TracCgi, having a slow response time for any page is expected there, as there's a lot of modules to load and initialization overhead that has to be repeated for each request (even more so if you misconfigured the rendering of static documents and use Trac's chrome module to render them instead of letting Apache do it - see TracCgi#MappingStaticResources).
You should really try to use either one of the faster deployment mode for Trac (basically anything but CGI, like TracModPython, TracModWsgi, TracStandalone).
You mentioned that you already tried with tracd, and that this didn't provide any performance gain. But can you try to monitor the tracd log output (with log_type = stderr and log_level = DEBUG, see TracLogging) and then try to figure out where tracd spends its time?
comment:8 by , 18 years ago
Dear, thanks for this great helpful response. I will try this these days and be back with some more debugging output ;). Thanks in advance…
comment:9 by , 18 years ago
I just enabled the debug output and got quite some warnings and errors. I will clean up everything and post here, if the performance stays that way…
comment:10 by , 18 years ago
ok, cleaning up the debug messages (resolving the sources of the warnings and errors) as well als moving to the apache (proxy) and tracd duo produced acceptable speed.
thanks for help
comment:11 by , 18 years ago
| Resolution: | → fixed |
|---|---|
| Status: | new → closed |
comment:13 by , 18 years ago
| Milestone: | 0.11.1 |
|---|---|
| Resolution: | → worksforme |
| Status: | reopened → closed |
So now that it works, if you don't mind you can also summarize what where those errors which made tracd very slow for you.
comment:14 by , 18 years ago
I'm not sure. I could not really determine what it made slow. I seems to me like that tracd is getting faster after about 1d of operating… before its as slow as the other stuff… there might be any caching reasons for…
comment:15 by , 17 years ago
| Component: | general → web frontend |
|---|---|
| Keywords: | cgi added |
comment:16 by , 16 years ago
| Resolution: | worksforme |
|---|---|
| Status: | closed → reopened |
Hi, I've installed trac 0.11.4 and it is too slow, unusable!!!! It makes around 15 seconds to load a page, more precisely all requests are granted and the page is entirily visible but it seems that there is something to wait and the load bar is completely loaded after 15 seconds. I use trac with apache2 and fastcgi. Is there a way to enhance performances? What can I do? Suggestions?
Thanks, in advance.
comment:17 by , 16 years ago
| Resolution: | → worksforme |
|---|---|
| Status: | reopened → closed |
Please don't re-open 15-month old tickets, especially if there is an open ticket on this very issue already (#7490).
Chances are that with only 512MB available the server is swapping, as there are known memory issues (see #6614). Can you move the server to a machine with more RAM? (at least 1GB). If not, you should at least use Python 2.5 which is somewhat better at giving memory back to the system.
A "100% load for about 30-45 second" is unfortunately not uncommon with lengthy pages like big changesets now that they're rendered with Genshi. There are ongoing efforts to speed things up at the Genshi level, including a GSoC proposal.
In order to be able to do something more specific about your issue, we need to know additional details: for which pages do you see this bad performance, all? some? If it's the timeline, you can reduce the default "daysback" parameter. For big changesets, there are other parameters to tune (TracIni#changeset-section), etc.