experiment: use a lazy bytestring hasql 'unknown'

[robx]

compare #2261

[robx]

Load test has no significant change. Memory use in mem-tests seems more or less significantly decreased:

< ok 1 - POST /rpc/leak?columns=blob: with a json key of 1M the memory usage(15,605,584 bytes) is less than 16M
> ok 1 - POST /rpc/leak?columns=blob: with a json key of 1M the memory usage(14,662,232 bytes) is less than 16M
< ok 2 - POST /leak?columns=blob: with a json key of 1M the memory usage(15,538,336 bytes) is less than 16M
> ok 2 - POST /leak?columns=blob: with a json key of 1M the memory usage(14,541,408 bytes) is less than 16M
< ok 3 - PATCH /leak?id=eq.1&columns=blob: with a json key of 1M the memory usage(15,585,384 bytes) is less than 16M
> ok 3 - PATCH /leak?id=eq.1&columns=blob: with a json key of 1M the memory usage(14,601,496 bytes) is less than 16M
< ok 4 - POST /rpc/leak?columns=blob: with a json key of 10M the memory usage(42,927,272 bytes) is less than 44M
> ok 4 - POST /rpc/leak?columns=blob: with a json key of 10M the memory usage(32,960,632 bytes) is less than 44M
< ok 5 - POST /leak?columns=blob: with a json key of 10M the memory usage(42,847,232 bytes) is less than 44M
> ok 5 - POST /leak?columns=blob: with a json key of 10M the memory usage(32,893,504 bytes) is less than 44M
< ok 6 - PATCH /leak?id=eq.1&columns=blob: with a json key of 10M the memory usage(42,900,040 bytes) is less than 44M
> ok 6 - PATCH /leak?id=eq.1&columns=blob: with a json key of 10M the memory usage(32,947,056 bytes) is less than 44M
< ok 7 - POST /rpc/leak?columns=blob: with a json key of 50M the memory usage(164,339,792 bytes) is less than 172M
> ok 7 - POST /rpc/leak?columns=blob: with a json key of 50M the memory usage(114,496,360 bytes) is less than 172M
< ok 8 - POST /leak?columns=blob: with a json key of 50M the memory usage(164,227,224 bytes) is less than 172M
> ok 8 - POST /leak?columns=blob: with a json key of 50M the memory usage(114,421,840 bytes) is less than 172M
< ok 9 - PATCH /leak?id=eq.1&columns=blob: with a json key of 50M the memory usage(164,282,832 bytes) is less than 172M
> ok 9 - PATCH /leak?id=eq.1&columns=blob: with a json key of 50M the memory usage(114,479,848 bytes) is less than 172M
< ok 10 - POST /perf_articles?columns=id,body: with a json payload of 32K that has 1000 array values the memory usage(12,605,744 bytes) is less than 14M
> ok 10 - POST /perf_articles?columns=id,body: with a json payload of 32K that has 1000 array values the memory usage(12,574,616 bytes) is less than 14M
< ok 11 - POST /perf_articles?columns=id,body: with a json payload of 329K that has 10000 array values the memory usage(13,508,120 bytes) is less than 14M
> ok 11 - POST /perf_articles?columns=id,body: with a json payload of 329K that has 10000 array values the memory usage(13,182,160 bytes) is less than 14M
< ok 12 - POST /perf_articles?columns=id,body: with a json payload of 3.4M that has 100000 array values the memory usage(22,804,512 bytes) is less than 24M
> ok 12 - POST /perf_articles?columns=id,body: with a json payload of 3.4M that has 100000 array values the memory usage(19,428,296 bytes) is less than 24M

[steve-chavez]

Memory use in mem-tests seems more or less significantly decreased:

Awesome! And this is by removing just one copy from what I understand, so it should be even better when removing more copies.

To ensure we don't increase on memory usage later, it'd be good to update the following tests:

postgrest/test/memory/memory-tests.sh

Lines 105 to 119 in fe8b32d

	jsonKeyTest "1M" "POST" "/rpc/leak?columns=blob" "16M"
	jsonKeyTest "1M" "POST" "/leak?columns=blob" "16M"
	jsonKeyTest "1M" "PATCH" "/leak?id=eq.1&columns=blob" "16M"
	jsonKeyTest "10M" "POST" "/rpc/leak?columns=blob" "44M"
	jsonKeyTest "10M" "POST" "/leak?columns=blob" "44M"
	jsonKeyTest "10M" "PATCH" "/leak?id=eq.1&columns=blob" "44M"
	jsonKeyTest "50M" "POST" "/rpc/leak?columns=blob" "172M"
	jsonKeyTest "50M" "POST" "/leak?columns=blob" "172M"
	jsonKeyTest "50M" "PATCH" "/leak?id=eq.1&columns=blob" "172M"
	postJsonArrayTest "1000" "/perf_articles?columns=id,body" "14M"
	postJsonArrayTest "10000" "/perf_articles?columns=id,body" "14M"
	postJsonArrayTest "100000" "/perf_articles?columns=id,body" "24M"

Load test has no significant change.

Hmm, I expected this to be affected too. Once this PR is merged I'll make some bench tests.

[robx]

Closing this one in favour of #2349 (which includes the same change via jsonBytes instead of unknown)