memory leak in gworldclock 1.4.4-9ubuntu1 (~450KiB/h)

[Selene ToyKeeper]

Binary package hint: gworldclock

I've discovered a memory leak in gworldclock in Ubuntu 11.04 (1.4.4-9ubuntu1). I measured it over time and found it leaks pretty steadily at a rate of about 450 to 500 KiB/h. I'll attach the graph below, but for now here are its config files:

> cat .gworldclock
<?xml version="1.0"?>
<gworldclock>
  <timeDisplayFormat>%Y-%m-%d %H:%M:%S (%a)</timeDisplayFormat>
</gworldclock>

> cat .tzlist
UTC "UTC"
America/Denver "Fort Collins"
America/Sao_Paulo "Diogo"
Europe/London "London"
Asia/Shanghai "Beijing"
Pacific/Auckland "Auckland"
Europe/Budapest "Budapest"

[Selene ToyKeeper (toykeeper)]

A very, very basic graph of memory use over time, sampled once per minute for about three days.

Memory use goes up by a small amount every few minutes, for an average of about 468KiB / hour.

[Matt Fischer (mfisch)]

Selene:

Is it 468KiB/hour as you said in comment #1 or per second as you said in the original description?

[Selene ToyKeeper (toykeeper)]

Ah, sorry. That was supposed to be per hour, but my fingers are used to typing "/s" instead. In any case, the graph shows 3 days of data and has clear units marked.

I just started a logger to check if this still happens in oneiric... will let you know the results.

[Matt Fischer (mfisch)]

Selene:

I ran it all night and check pmap every minute. I did not see an increase other than one initial bump. It's been here since about 9pm last night:

Mon Dec 5 08:42:57 MST 2011 - total 331460K

This is in oneiric.

I used your same .tzlist file. Let me know what you see.

[Selene ToyKeeper (toykeeper)]

I ran it for ~5 hours and it increased in size from about 13.25 MiB to 15.45 MiB, which is about 440 KiB / hour. When graphed, it makes a rather straight diagonal line with no interesting features after the first 5 minutes or so.

To measure, I'm just running 'ps axu' once per minute, grepping for the process, and dumping the time and RSS value to a file. I think pmap measures VSZ instead of RSS, which may not produce useful results. Can you try it with an RSS measurement to verify whether it happens outside my environment?

I'm using gworldclock 1.4.4-9ubuntu1 on oneiric x86_64... which seems to be the same version that natty had. I suspect apt didn't even touch it during my dist-upgrade.

[Matt Fischer (mfisch)]

So I confirmed this, although my rate is closer 240k/hour. I will work on sending this upstream.

mfisch@caprica:~$ while true; do ps axu | grep gworldclock | grep -v grep; sleep 300; done
USER PID %CPU %MEM VSZ RSS TTY STAT START TIME COMMAND
mfisch 22671 1.0 0.1 318916 12964 pts/0 Sl 11:55 0:12 gworldclock
mfisch 22671 1.0 0.1 319076 13024 pts/0 Sl 11:55 0:15 gworldclock
mfisch 22671 1.0 0.1 319076 13024 pts/0 Sl 11:55 0:18 gworldclock
mfisch 22671 1.0 0.1 319076 13028 pts/0 Sl 11:55 0:21 gworldclock
mfisch 22671 1.0 0.1 319076 13032 pts/0 Sl 11:55 0:25 gworldclock
mfisch 22671 1.0 0.1 319076 13032 pts/0 Sl 11:55 0:28 gworldclock
mfisch 22671 1.0 0.1 319076 13036 pts/0 Sl 11:55 0:30 gworldclock
mfisch 22671 1.0 0.1 319076 13036 pts/0 Sl 11:55 0:33 gworldclock
mfisch 22671 1.0 0.1 319076 13036 pts/0 Sl 11:55 0:36 gworldclock
mfisch 22671 1.0 0.1 319076 13036 pts/0 Sl 11:55 0:39 gworldclock
mfisch 22671 1.0 0.1 319076 13164 pts/0 Sl 11:55 0:43 gworldclock

[Matt Fischer (mfisch)]

Thanks for your report. Since this is an unmodified copy of debian's code, I sent it upstream to the maintainer there:

http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=651129