memory error

[DaveNZ]

Running build 955, got a bad error come up - something about running out of memory.
It made a .elf file and told me to report it...but the forum won't let me upload it?

Also note that after this happened Cumulus could not communicate with the weather station again until the computer was restarted.

[DaveNZ]

ok, turns out .elf file is really just a txt file...

[DaveNZ]

Cumulus has been running for 1 hour now and in that time the RAM Cumulus is using has gone from under 24,000K to 37,500K.
I hope it's not heading down the same path (and crashes again)
If it does I will reinstall build 954 for the time being

[steve]

Possibly a memory leak somewhere. Mine is up to 33MB, having been running for about 11 hours, but doesn't seem to be getting any higher. Had you been doing anything in particular in that hour - viewing graphs etc?

I've added .elf as a permitted attachment.

Edit: Looking at the dump, it's the debug log causing the problem - have you been running with it turned on (for the graphs problem or something?), and the file has got very big? It's best to delete the debug log from time to time.

[DaveNZ]

I had deleted the debug log 2 hours earlier (it was 2MB) so wouldn't think it was that.
I often view the graphs, and have the graph window left open at times during the day

I have the RAM increasing without bound it seems - might be the ftplog that you said it is writing even though it's disabled?
Memory usage hit around 70MB before I closed Cumulus and deleted the ftplog.
Shortly after restarting memory usage was 22MB...back up to 50MB now

[steve]

might be the ftplog that you said it is writing even though it's disabled?

Possibly, the build I'm using doesn't have that problem now, but I've turned ftp logging on and it doesn't happen. I've tried a number of things and I can't get it to happen.

If you restart it and don't touch it, and watch it in task manager, does the memory usage still increase in the same way?

[DaveNZ]

Ok have restarted and won't do anything in Cumulus. Straight after restart its on 14MB memory.
25 minutes after my last post it had crept from 50MB to 60MB

Will post something in 1-2hrs to see how much its increased

[steve]

One thing occurs to me - a major change in 955 is that the web tag handling is completely rewritten. If you're using a particular web tag that I'm not, particularly for realtime ftp, then that might be the cause of the leak.

[DaveNZ]

After following your suggestion of not viewing any graphs or anything, the memory use has risen very slowly to 18MB over 1hr 20 mins, which is probably pretty normal. What does this mean?

[steve]

I've found (and fixed) a very small leak each time VP2 data is read, which probably accounts for your slow increase.

As for the large increase, you'll need to experiment to see what causes it. Try just one thing at a time. I've tried viewing graphs, zooming them, ftp logging on and off, debug logging on and off, etc etc with no increase. I looked at the new web tag handling and there's nothing there that could cause a leak.

[DaveNZ]

After I brought up the graphs window the memory went from 18MB to 26MB.

Now I've found the pattern - in the graphs window, every time I click a thumbnail the memory jumps up about 0.9MB - but this jump only occurs while a web update is underway.

I've managed to bring the memory to 44MB repetitively clicking thumbnails while a web update is happening

[DaveNZ]

Actually I lie - just opened graph window and clicked thumbnails and got memory jumps again - no web update happening though. Clicking thumbnails that were very recently clicked on has no effect though.

All I can think of is maybe every time I view an updated graph for the first time, the memory increases. Anyway, something to scratch your head over!

[steve]

Are you sure about that? I can make it use more memory during an upload, as you said, but it doesn't use any more just after an update (on the minute). I've managed to make it (the chart component) crash too, with all this clicking...

What seems to happen is that when you view a graph, more memory is used, which I suppose is fair enough, and then after some other event (like creating the image for upload) viewing the graph again uses more memory.

[daj]

It does seem to be related to the graphs/web updating.

My Cumulus (1.9.1 b955) was running for about the last 12 hours un-touched and was using around 31Mb memory.

I then went into the Graphs and clicked on each one, which added around 6Mb memory. Closing the Graphs window down and going back in to move around the graphs does not increase memory (well, maybe a few Kb).

If I force a web update and then go back into the graphs and move around it adds another 6Mb -- and does this each time I have a web update and then look at graphs.

I have no idea if this was doing it in earlier versions of Cumulus as I do not jump in and out of graphs too often, so as potential memory leaks go it is not that big a deal

[DaveNZ]

Are you sure about that? I can make it use more memory during an upload, as you said, but it doesn't use any more just after an update (on the minute). I've managed to make it (the chart component) crash too, with all this clicking...

I'm pretty sure, I watched the memory jump up in the situation with no upload and with an upload happening. By no upload I mean no proper web update, but realtime.txt is still uploading every 10 seconds.

What seems to happen is that when you view a graph, more memory is used, which I suppose is fair enough, and then after some other event (like creating the image for upload) viewing the graph again uses more memory.

Yes but how come after the graph is viewed and the graph window is closed, the memory does not decrease?

I seem to have more than my fair share of anomalies - I protest!