Finding memory leaks(gent.ilcore.com)
gent.ilcore.com
Finding memory leaks
http://gent.ilcore.com/2011/08/finding-memory-leaks.html
4 comments
It's too easy for the web developers to just point the memory
leak blame to the browser and none of the users will be the wiser.
Nowadays, garbage collection in behaves much better. Even though browsers possibly consume more monstrous amounts of memory than ever before, they tend to give much of it back when the corresponding pages are closed. It's time to bring web devs back to the same table they were told to stay away from only a few years ago. But don't be surprised if this takes a bit longer.
If users start complaining about the performance of a site, and memory leaks are the cause, web developers will start fixing memory leaks. It's easy for them to initially say, "it's the browser's fault," but eventually, someone is going to start poking around in the site itself if they get enough complaints.
By the way, I'm not sure what architecture other browsers use, but Chrome already has a separate process for each tab. Under the wrench icon in the far right, if you go to Tools -> Task Manager, you can see the memory used by each tab.
By the way, I'm not sure what architecture other browsers use, but Chrome already has a separate process for each tab. Under the wrench icon in the far right, if you go to Tools -> Task Manager, you can see the memory used by each tab.
I think that that day is rapidly approaching or here already. Using chrome you can inspect per tab memory usage and use that to identify leaks that may be coming from a page (in fact that's what the author did originally). And in firefox you can use about:memory to inspect per tab JS memory compartments. Its still not always obvious who is to blame (browser or site), but it is becoming very easy to spot this kind of behavior.
For anyone trying this at home: per-compartment memory reporting is a new feature of about:memory in Firefox 7. You'll need a build from the Aurora channel to try it out right now: http://firefox.com/channel
Back in the good old days, you would actually run out of memory. It kept you on your toes unlike today when allocs rarely if never fail. Instead, everything just collectively gets paged out to disk, then sluggish and sloshed, and eventually inoperable. And at that point you really can't pinpoint it down to some stupid web page because everything is equally paged out by the time.
THe browser could turn memory hogging tabs to red. It could impose an absolute or relative limit to each tab's memory usage. Or what used to be nice in the old days, having each tab expose a function that causes it to minimize its memory usage, i.e. ditch any runtime data that can be recomputed later. But since webpages are programs these days, this interface ought to be negotiated between the browser and creators of web pages. Not going to happen.
Since web pages are sort of programs, I like the Chrome way of putting them into separate processes. Then let the operating system manage them: it's much better suited for that than a browser. Using setrlimit(2) to limit the amount of memory could be useful. Automatic killing and restarting of a web page upon hitting these limits would give an incentive for the developers to keep their pages small; or not keep them small but design them to just deal with these mandatory resets.
THe browser could turn memory hogging tabs to red. It could impose an absolute or relative limit to each tab's memory usage. Or what used to be nice in the old days, having each tab expose a function that causes it to minimize its memory usage, i.e. ditch any runtime data that can be recomputed later. But since webpages are programs these days, this interface ought to be negotiated between the browser and creators of web pages. Not going to happen.
Since web pages are sort of programs, I like the Chrome way of putting them into separate processes. Then let the operating system manage them: it's much better suited for that than a browser. Using setrlimit(2) to limit the amount of memory could be useful. Automatic killing and restarting of a web page upon hitting these limits would give an incentive for the developers to keep their pages small; or not keep them small but design them to just deal with these mandatory resets.
> Back in the good old days, you would actually run out of memory. It kept you on your toes unlike today when allocs rarely if never fail. Instead, everything just collectively gets paged out to disk, then sluggish and sloshed, and eventually inoperable. And at that point you really can't pinpoint it down to some stupid web page because everything is equally paged out by the time.
There's always swapoff(8) if you want that behavior back...might want to tweak your /proc/$PID/oom_adj (assuming Linux) to make sure the OOM-killer axes the "right" process, though.
There's always swapoff(8) if you want that behavior back...might want to tweak your /proc/$PID/oom_adj (assuming Linux) to make sure the OOM-killer axes the "right" process, though.
this is the type of article I enjoy seeing on HN.
This is a great article.
A good portion of my day to day work involves handling JS memory leaks, due to the growing reliance on JS for behaviours, interactions, and application logic (sometimes an over-reliance, IMHO). It's sad that most of the JavaScript developers I interview have no real idea what causes or how to fix leaks. We could really do with making the community more aware of this stuff!
A good portion of my day to day work involves handling JS memory leaks, due to the growing reliance on JS for behaviours, interactions, and application logic (sometimes an over-reliance, IMHO). It's sad that most of the JavaScript developers I interview have no real idea what causes or how to fix leaks. We could really do with making the community more aware of this stuff!
Talking to JS programmers about memory management is like talking to VisualBasic programmers about data alignment. Won't lead anywhere.
Fixing leaking websites should be done by the JS vm as webdevs are usually pretty non technical and have no understanding of concepts like memory management.
Fixing leaking websites should be done by the JS vm as webdevs are usually pretty non technical and have no understanding of concepts like memory management.
It's too easy for the web developers to just point the memory leak blame to the browser and none of the users will be the wiser.
If some day in the future there's an extremely easy way to figure out which sites are using how much memory (each tab in a process?) then we could start seeing some push back for JS leaks to be fixed.