Fix timer GC delays in the Linux filesystem collector

[jordy1024]

Hello,I found a potential hazard in filesystem_linux.go that may cause memory leaks,the details are as follows:
function GetStats will loop through mount points, according to the number of all mount points.
Recently I encountered a problem: the number of mount points of a certain system in the production environment is abnormal, as many as more than 500,000 rows; mainly because the for loop is blocked, and then the Timer Heap objects in the statement for {select case->time.After cause accumulation, and eventually a memory leak occurs.
It may be a better way to control the number of mount points to be checked, but this may require more changes; therefore, I only made a slight optimization for this problem: change time.After to time.NewTimer.
Expect to be accepted or answered.

[discordianfish]

Hi,

not sure I understand the implications of this change. If you have 500k mountpoints, having 500k timer objects that disappear again after the timeout is expected. And from what I can tell, with your change this would still be the case, right?

[jordy1024]

@SuperQ hi, why ci/circleci report "Your tests failed on CircleCI" ？ What should I do ?

[SuperQ]

I think the issue with CircleCI is because you've selected to "Follow" the node_exporter fork in your CircleCI account. This triggers CircleCI to think that it should build under your account, rather than the Prometheus account.

If you click the "Unfollow Project" for node_exporter in your CircleCI account, we can try to re-trigger builds.

[jordy1024]

I think the issue with CircleCI is because you've selected to "Follow" the node_exporter fork in your CircleCI account. This triggers CircleCI to think that it should build under your account, rather than the Prometheus account.

If you click the "Unfollow Project" for node_exporter in your CircleCI account, we can try to re-trigger builds.

Okay, maybe this is my problem;
When I saw a CI error, I immediately clicked and performed the related sign up, and also clicked follow, haha ^_^.
It has been cancelled(Unfollow Project), so I don’t need to do anything for the next build step?

[SuperQ]

You might need to make a commit to trigger the build. Maybe squash up the commit history with git rebase -i origin/master?

[jordy1024]

You might need to make a commit to trigger the build. Maybe squash up the commit history with git rebase -i origin/master?

Okay,Thanks.
Please review again, looking forward to being approved.

[SuperQ]

LGTM, for the record if you read the Go docs for time.After() this is explicitly mentioned.

After waits for the duration to elapse and then sends the current time on the returned channel. It is equivalent to NewTimer(d).C. The underlying Timer is not recovered by the garbage collector until the timer fires. If efficiency is a concern, use NewTimer instead and call Timer.Stop if the timer is no longer needed.

So, yes, it's much better to explicitly call Stop() when there's nothing to wait for anymore.

[roidelapluie]

Awesome findings!