Disable key fields check if unnecessary and optimize its perf #3720
Conversation
| /** | ||
| * Returns whether the `typePolicy` directive is present on at least one object in the schema | ||
| */ | ||
| fun hasAnyTypePolicyDirectives(): Boolean { |
There was a problem hiding this comment.
Nit: rename to hasTypeWithTypePolicy ?
|
|
||
| private val keyFieldsCache = mutableMapOf<String, Set<String>>() | ||
| fun keyFields(name: String) = keyFieldsCache.getOrPut(name) { | ||
| schema.keyFields(name) |
| parentType: String, | ||
| implementedTypes: Set<String>, | ||
| ) = collectFieldsCache.getOrPut("$selections $parentType $implementedTypes") { | ||
| collectFields(selections, parentType, implementedTypes) |
There was a problem hiding this comment.
Do you have any idea the cache hit ratio for this? I'd expect it to be relatively low?
There was a problem hiding this comment.
With the test in the PR I can see that the method was run 2693 times overall, and only 405 times after making it cached. So 2288 duplicated calls, meaning ~80% cache hit?
| checkKeyFields(it, options.schema, allFragmentDefinitions) | ||
| // Check if all the key fields are present in operations and fragments | ||
| // (do this only if there are key fields as it may be costly) | ||
| if (options.schema.hasAnyTypePolicyDirectives()) { |
| import kotlin.time.measureTime | ||
|
|
||
| @OptIn(ExperimentalTime::class, ApolloExperimental::class) | ||
| class PerfTest { |
There was a problem hiding this comment.
Could we do everything from CodegenTest ? CodegenTest writes a measurements file containing the codegen time IIRC so it might be enough?
There was a problem hiding this comment.
Yeah I really don't think we should push that test and indeed it should be one of the regular tests ran by CodegenTest (in fact it already does because where the files are located, and fails because I didn't generate the .expected 馃槄). But since it's fairly slow, and we may want to continue investigations, it's easier to have a dedicated test temporarily. If we want to merge this branch, let's remove or move it first.
There was a problem hiding this comment.
Would be nice to merge the other fixes quickly. My moving the test to another PR?
BoD
force-pushed
the
check-key-fields-optimize
branch
from
6007e4b
to
8eaa0d6
Compare
BoD
changed the title
| * Returns whether the `typePolicy` directive is present on at least one object in the schema | ||
| */ | ||
| fun hasTypeWithTypePolicy(): Boolean { | ||
| return typeDefinitions.values.filterIsInstance<GQLObjectTypeDefinition>().any { objectType -> |
There was a problem hiding this comment.
Note for our future selves: we'll want to support @typePolicy on interfaces and unions (see #3356). This will need to be updated when this happens
This branch investigates a performance issue in the compiler around check key fields, that has been reported.
typePolicydirectivesimplementedTypesandkeyFieldscollectFieldsentirely馃憜 when profiling, these changes seem to show the check runs faster, however we still don't have a convincing repro steps where it was taking a significant time overall, so caution must be taken