STRLCPY/graphql-engine

server: remove default `+RTS -qn2` setting

We notice significant improvements to max GC pause time, for huge schema (both on 8.10 and 9.2)

This was added when we were on ghc-8.6.5. The referenced issues (from old oss repo) seem irrelevant or don't have useful information afaict

re. https://hasurahq.atlassian.net/browse/SRE-10
https://hasurahq.atlassian.net/browse/SRE-129

-----

**EDIT**: old discussion https://hasurahq.slack.com/archives/CKFUG6RCH/p1576518775017100

It sounds as though if the `replace_metadata` benchmark doesn't regress then we're okay

**EDIT**: Ooh nice, replace_metadata is much faster across the benchmark sets (see "both compared" details)

PR-URL: https://github.com/hasura/graphql-engine-mono/pull/6133
GitOrigin-RevId: a9ca4d8a25cce06a6fa4e85ce36cfd6f6c2bb0cd

Brandon Simmons committed with hasura-bot 2 months ago

a9202c42

1 parent f7bef061

Total 1 files

■ ■ ■ ■ ■ ■

server/graphql-engine.cabal

		skipped 119 lines
120	120		-- `-I0` disables idle GC. We do this in application code now. See 'ourIdleGC' for details.
121	121		-- `-T` is required to collect the stats we use in 'ourIdleGC'.
122	122		--
123		-	-- `-qn2` limits the parallel GC to at most 2 capabilities. This came up in #3354/#3394, as the
124		-	-- parallel GC was causing significant performance overhead. Folks in #ghc on freenode advised
125		-	-- limiting the parallel GC to 2 or 3 capabilities as a very conservative choice, since more
126		-	-- than that is highly unlikely to ever be helpful. More benchmarking would be useful to know if
127		-	-- this is the right decision. It’s possible it would better to just turn it off completely.
128		-	--
129	123		-- `-kc8K` helps limit memory consumption in websockets (perhaps elsewhere) by making the
130	124		-- cost of a thread's first (and probably only) stack overflow less severe.
131	125		-- See:https://github.com/hasura/graphql-engine/issues/5190
		skipped 1 lines
133	127		-- `--disable-delayed-os-memory-return` seems to help lower reported residency, in particular
134	128		-- in situations where we seem to be dealing with haskell heap fragmentation. This is more a
135	129		-- workaround for limitations in monitoring tools than anything...
136		-	"-with-rtsopts=-N -I0 -T -qn2 -kc8K --disable-delayed-os-memory-return"
	130	+	"-with-rtsopts=-N -I0 -T -kc8K --disable-delayed-os-memory-return"
137	131
138	132		common lib-depends
139	133		build-depends: Spock-core
		skipped 1211 lines