Possible memory leak in ssh.exec about ssh2

[mscdex]

What else would you expect? You're creating new channel streams every second, regardless of whether or not the callback has been executed and the 'exit' event has been emitted on the stream?

[VadimIpatov]

Unfortunately, destruction of streams doesn't prevent memory leak. For example: https://dl.dropboxusercontent.com/u/3261328/testssh2.js

What should I do in order to execute a command every second using a single ssh connection?

[mscdex]

I would suggest wrapping your exec() call into a separate function. Call it immediately on 'ready' and then on the stream's 'exit' event, do a setTimeout(doExec, 1000);.

I would also suggest using some profiling tools to verify that there is in fact a leak and if so, where it is actually located.

[VadimIpatov]

Thank you! I'll try to do as you suggest.

[VadimIpatov]

Unfortunately, it didn't help.
I tried to run this code: https://dl.dropboxusercontent.com/u/3261328/testssh2_setTimeout.js
Here the results: https://dl.dropboxusercontent.com/u/3261328/testssh2_results.txt

Something like ssh.exec('watch -n1 cmd', function(err, stream) {...} ) runned once don't cause leakage.

[mscdex]

setTimeout(doExec(), 1000); should be setTimeout(doExec, 1000);

EDIT: Additionally, you cannot expect it to execute exactly every second due to server latency and especially the routinely forced garbage collection.

[VadimIpatov]

Oh, it's my mistake...
But the memory leak is still present: https://dl.dropboxusercontent.com/u/3261328/testssh2_results2.txt

you cannot expect it to execute exactly every second

It's not a big problem. In my task such precision is not important. And manual GC will be turned off in production.

[VadimIpatov]

Any ideas?

[VadimIpatov]

What do you think, it's a bug or my mistake?
I see the only solution at the moment: use the watch -n 1

[stinger1206]

I have the same problem.

[mscdex]

After much digging (special thanks to heapdump) and investigation, I found the leak. It should now be plugged in the master branch.

Can both of you give it a try?

[VadimIpatov]

Oh, thank u! It's much better now.
Unfortunately, leak is still there, but not so fast as before.
There is my tests: https://dl.dropboxusercontent.com/u/3261328/testssh2_results3.txt

[mscdex]

@VadimIpatov Let it run for much longer. I have been running it for several hours now (with manual full GCs to more easily ensure no leaks are present). During this time, it has been hovering around 3.17mb for the heap (process.memoryUsage().heapUsed) and 20mb for the rss (process.memoryUsage().rss).

[VadimIpatov]

Yeah, you're right. It works! :)
Thank you!