STRLCPY/graphql-engine

benchmarks: Add `Accept-Encoding: gzip` headers (NO REGRESSION)

PR-URL: https://github.com/hasura/graphql-engine-mono/pull/5622
GitOrigin-RevId: d4952b00cfe322f9f80d1005412ac3c777f3bd46

Brandon Simmons committed with hasura-bot 2 months ago

d6703811

1 parent 4a582532

Total 5 files Show one by one

■ ■ ■ ■ ■ ■

server/benchmarks/README.md

		skipped 85 lines
86	86		https://well-typed.com/blog/2021/01/fragmentation-deeper-look/
87	87		https://well-typed.com/blog/2021/03/memory-return/
88	88
	89	+	- If optimizing or tuning the output/compression codepath:
	90	+	- `chinook`.`_small_result` and `simple_query_` queries are average wrt
	91	+	response body size (according to cloud data)
	92	+	- ...and `chinook`.`full_introspection` is ~P95
	93	+
89	94		## Adding a new benchmark and reviewing
90	95
91	96		You'll create a new directory under `benchmark_sets/`, and in general can
		skipped 42 lines

■ ■ ■ ■ ■ ■

server/benchmarks/benchmark_sets/chinook/config.query.yaml

		skipped 4 lines
5	5		X-Hasura-Admin-Secret: my-secret
6	6		X-Hasura-Role: employee
7	7		X-Hasura-User-Id: 4
	8	+	# On cloud about 95% of requests incoming Accept gzip (which as of writing
	9	+	# means we unconditionally compress, which has significant performance
	10	+	# impact), so it's important that we include it in all benchmarks unless we
	11	+	# have an intentional reason not to.
	12	+	Accept-Encoding: gzip
8	13
9	14		# Anchors to help us DRY below; settings here may be overridden selectively
10	15		constants:
		skipped 25 lines
36	41		############################################################################
37	42		# single-table query, small result size; makes use of permissions for
38	43		# filtering; low RPS
	44	+	#
	45	+	# NOTE: According to data from cloud this is a pretty average response (we see
	46	+	# P50 compressed response body size of 123 bytes, and P75 290 bytes):
	47	+	#
	48	+	# uncompressed body size: ~600 bytes
	49	+	# compressed body size: ~200 bytes (as of time of writing)
39	50		- name: simple_query_low_load
40	51		<<: *k6_custom
41	52		options:
		skipped 40 lines
82	93
83	94
84	95		######## Small result size
	96	+	#
	97	+	# uncompressed body size: ~650 bytes
	98	+	# compressed body size: ~200 bytes
	99	+
85	100		- name: complex_query_low_load_small_result
86	101		<<: *k6_custom
87	102		options:
		skipped 49 lines
137	152
138	153
139	154		######## Large result size
	155	+	#
	156	+	# NOTE: According to data from cloud, this is somewhere between
	157	+	# P90 (2 kB compressed) and P95 (4 kB compressed):
	158	+	#
	159	+	# uncompressed body size: ~33 kB
	160	+	# compressed body size: ~3 kB (as of time of writing)
	161	+
140	162		- name: complex_query_low_load_large_result
141	163		<<: *k6_custom
142	164		options:
		skipped 32 lines
175	197		# The standard introspection query from server/src-rsr/introspection.json
176	198		# We don't expect a server to be hammered with these, but they are crucial
177	199		# for clients (like graphiql):
	200	+	#
	201	+	# NOTE: According to data from cloud, this is somewhere between
	202	+	# P95 (4 kB compressed) and P99 (40 kB compressed):
	203	+	#
	204	+	# uncompressed body size: ~190 kB
	205	+	# compressed body size: ~13 kB (as of time of writing)
178	206
179	207		- name: full_introspection
180	208		<<: *k6_custom
		skipped 110 lines

■ ■ ■ ■ ■ ■

server/benchmarks/benchmark_sets/huge_schema/config.query.yaml

1	1		# This tells graphql-bench that it's testing a hasura instance and should
2	2		# collect some additional metrics:
3	3		extended_hasura_checks: true
4		-	# headers:
5		-	# X-Hasura-Admin-Secret: my-secret
	4	+	headers:
	5	+	Accept-Encoding: gzip
6	6
7	7		# Anchors to help us DRY below; settings here may be overridden selectively
8	8		constants:
		skipped 310 lines

■ ■ ■ ■ ■ ■

server/benchmarks/benchmark_sets/remote_schema/config.query.yaml

1	1		# This tells graphql-bench that it's testing a hasura instance and should
2	2		# collect some additional metrics:
3	3		extended_hasura_checks: true
4		-	# headers:
5		-	# X-Hasura-Admin-Secret: my-secret
	4	+	headers:
	5	+	Accept-Encoding: gzip
6	6
7	7		# Anchors to help us DRY below; settings here may be overridden selectively
8	8		constants:
		skipped 230 lines

■ ■ ■ ■ ■ ■

server/src-lib/Hasura/Server/Compression.hs

		skipped 41 lines
42	42		contentEncodingHeader :: CompressionType -> NH.Header
43	43		contentEncodingHeader CTGZip = gzipHeader
44	44
45		-	-- \| Maybe compress the response body, based on the client's Accept-Encoding
46		-	-- and our own judgement.
47	45		compressResponse ::
48	46		NH.RequestHeaders ->
49	47		BL.ByteString ->
		skipped 171 lines