Log.CloseAndFlush() deadlock on AWS Lambda

[nikoudel]

Log.CloseAndFlush() seems to be deadlocking when run in AWS Lambda and my Seq server instance goes down for some (irrelevant) reason making Seq a single point of failure for the whole microservice gang. This is how it looks like in the logs:

This only happens when running in AWS Lambda environment. In a console app execution continues after a graceful timeout of ~4 seconds.
I suspect it might be related to this and followed this advice to make it behave better:

...
finally
{
    Task.Run(Log.CloseAndFlush).Wait(1000);
}

It works but still looks like a hack.

[nblumhardt]

Thanks for the report, we'll dig in 👍

[DarkMio]

We're affected by the same problem. CloseAndFlush causes in non-debug environments to deadlock. One of our engineers pointed out that a lock and tasks cause tasks running in a sync context - which then deadlocks by accident.

This only happens with a config in which the seq sink is added, the others we use (console, eventlog) are unaffected.

[nblumhardt]

Hi @DarkMio - thanks for reaching out. Starting work on a proper fix for this now via serilog/serilog#1750.

[DarkMio]

Very cool, that was just in time. Thanks @nblumhardt!

[nblumhardt]

Hi @DarkMio; v5.2.0 is out now and includes this fix. I'll close the ticket as completed but if you notice any ongoing issues please let us know!

[nblumhardt]

Ah sorry, also, for this to work you'll need to asynchronously dispose the Serilog Logger. If you're using var logger = new LoggerConfiguration().CreateLogger() that maps to await using; otherwise, using the Log class it's await Log.CloseAndFlushAsync().

If you're on Serilog.AspNetCore and configuring Serilog in the UseSerilog(lc => ...) callback, you'll need:

serilog/serilog-extensions-hosting#67