do not waste time when metrics delegate is not set

[weissi]

Motivation:

Unfortunately, even if no metics delegate has been configured, significant work is being done each EL tick. Concretely:

• 2 NIODealine.now() calls which are expensive particularly on some hypervisors (may cause a VMEXIT!)
• Some unnecessary calculations

Modifications:

• Only perform the metrics delegate work if a delegate has actually been set

Result:

If no delegate is set, each EL tick will now be faster.

[weissi]

CC @hamzahrmalik

[glbrntt]

Thanks!

[weissi]

@glbrntt do you know about the benchmarks?

--- a/Benchmarks/Thresholds/6.2/NIOPosixBenchmarks.TCPEcho.p90.json
+++ b/Benchmarks/Thresholds/6.2/NIOPosixBenchmarks.TCPEcho.p90.json
@@ -1,3 +1,4 @@
 {
-  "mallocCountTotal" : 608
-}
+  "mallocCountTotal" : 613,
+  "wallClock" : 6097356667
+}
\ No newline at end of file
diff --git a/Benchmarks/Thresholds/6.2/NIOPosixBenchmarks.TCPEchoAsyncChannel.p90.json b/Benchmarks/Thresholds/6.2/NIOPosixBenchmarks.TCPEchoAsyncChannel.p90.json
index f0d21b9..b0c41ef 100644
--- a/Benchmarks/Thresholds/6.2/NIOPosixBenchmarks.TCPEchoAsyncChannel.p90.json
+++ b/Benchmarks/Thresholds/6.2/NIOPosixBenchmarks.TCPEchoAsyncChannel.p90.json
@@ -1,3 +1,4 @@
 {
-  "mallocCountTotal" : 82530
+  "mallocCountTotal" : 82278,
+  "wallClock" : 14359068200
 }

what does that mean?

[Lukasa]

Nothing, it means that swift-package-benchmark has stopped being reliable. I'll re-run for now.

[weissi]

Nothing, it means that swift-package-benchmark has stopped being reliable. I'll re-run for now.

Hmm, I've done it a bunch of times for 6.2, seems consistently failing. But I did only re-run the failing tests not everything

[weissi]

@Lukasa yeah failed again :\

[weissi]

@Lukasa how can I get this in? The PRBs are very unstable, sometimes 1 fails, sometimes 3 fail, never gotten it to zero :'(

[glbrntt]

@Lukasa how can I get this in? The PRBs are very unstable, sometimes 1 fails, sometimes 3 fail, never gotten it to zero :'(

The alloc count is low in that test (TCPEcho) and presumably amortises to zero per iteration so we could just add some slack to that particular test. There's an example in the TCPEchoAsyncChannel test of how to do this. WDYT?

[Lukasa]

Works for me.