Hangs NodeJS on shutdown when using maxAge.

[lscpike]

NodeJs 0.10.33

When I shut down the server it hangs up to the maxAge time before the process exits. Suggests the cache expiry timer is keeping the event loop going.

If I clear the cache explicitly the process exits immediately, but this means I need to track all memoized functions.

[medikoo]

I see the issue. How do you shut down the server? Just programmatically within a process, or do you send some signal to the process, and server shuts down, but timeouts keep it going?

[lscpike]

Handling the SIGINT or SIGTERM signal for example. A debugger sends a SIGINT on stop I'm pretty sure. Basically any method that allows the process to finish processing current requests/work.

In this particular case I call the shutdown method on a custom server so it can record state and log4js so it can flush to disk.

[medikoo]

@lscpike thanks, so solution is easy, I'll make sure all timeouts are cleared on kill signal.

[lscpike]

That might work fine, but I'm not sure whether the one signal is enough. A more precise solution might be to call unref on the timeoutObject returned from setTimeout. However, the docs suggest this might cause performance issues if used too often.

[iammaart]

How about making 'unreffing' of the timers created by maxAge an extra configuration option? That way the user can decide whether the potential performance penalty is worth it.

[nevercast]

I just observed this, currently I'm reducing maxAge to a time I'm willing to wait for the process to exit.

[puzrin]

Not sure that listening global kill is the best idea. You may not know how user handle additional signals like SIGHUP/SIGUSR1/SIGUSR2. For example, he may wish to do soft restart of workers in cluster. I'd vote for global method to clear all timeouts (if that possible). That can be useful when soft restart needed.

[medikoo]

@puzrin I agree, also It may be totally different case on Windows.

[puzrin]

Found: https://nodejs.org/api/timers.html#timers_unref

Timers in memoizee should be "unref-ed". That will do exactly what we need.

[medikoo]

@puzrin indeed, great found! We'll follow that then

[nevercast]

Not sure if the bug is still around but unref used to cause memory leaks.
I know the npm:odbc-pool module has a .drain() method that clears all the timeouts for use when shutting down your application. That might be some reference.

unref is likely a suitable option, as it's meant to work.

[puzrin]

In the case of setTimeout when you unref you create a separate timer that will wakeup the event loop, creating too many of these may adversely effect event loop performance -- use wisely.

Problem could be in creating too many timeouts. I noted somewhere in another issues, that timer ticks management should be done by single timer object.

[fampinheiro]

I'm currently hitting this problem. Do you settled in a solution?

[medikoo]

@fampinheiro no it's still not addressed, It will for sure be addressed for v1, but date when that happens is uncertain. PR's are welcome.

[madnight]

Same problem here under Arch Linux. The maxAge values prevents node from shutdown for the time period defined in maxAge.

[medikoo]

Fixed following advice by @puzrin from #25 (comment) (not sure why it took me so long, sorry about that).

Published as v0.4.12

[medikoo]

On additional note, with v1 there's a plan to not rely on timers at all.
Whether value is stale will be confirmed at invocation. Downside is that cache may grow in memory without control, but that can be further adjusted with max setting