Absinthe: Obscenely high memory usage with deeply nested inputs during Absinthe.Blueprint.Transform.walk
Environment
- Elixir version (elixir -v): 1.6.1
- Absinthe version (mix deps | grep absinthe): 1.4.12
- Client Framework and version (Relay, Apollo, etc): N/A
Expected behavior
Low memory usage, and a message informing me that my input type was invalid or my mutation was invalid.
Actual behavior
Memory usage on the beam process ballooned to 4GB fairly rapidly, before the test timed out - in production, though, this broke the system and we had to restructure how this mutation was written.
It seems to strictly be a function of the level of nesting in input objects, at first glance.
Relevant Schema/Middleware Code
Reproduction here: https://github.com/knewter/absinthe_memleak_reproduction
Relevant test (running this against any schema whatsoever should reproduce the issue):
defmodule AbsintheMemleakReproductionTest do
use ExUnit.Case
doctest AbsintheMemleakReproduction
test "deeply nested inputs" do
# If I'm right, this will lead to a massive memory leak when trying to walk the blueprint
query = """
mutation f($input: SomeInput!){
doThing(input: $input){
id
}
}
"""
variables = %{
"input" => %{
"a" => "1",
"b" => "1",
"c" => "1",
"d" => "1",
"e" => "1",
"f" => "1",
"g" => [
%{
"h" => [
%{
"i" => %{
"bb" => [
%{
"cc" => [
%{
"dd" => [
%{
"l" => [
%{
"m" => "fa623f91-ed57-4d2c-b7e4-e67371ad5b23",
"n" => [
%{
"ii" => "ada67902-4a11-4594-afc3-7c912b445080",
"jj" => [
%{
"kk" => "0b65185e-06bb-466e-b213-ba18e9ae2135",
"o" => "Colombia",
"gg" => "0.0"
}
],
"o" => "Coffee Variety"
},
%{
"ii" => "82e42b16-c79f-472a-b109-38c3d529e27d",
"jj" => [
%{
"kk" => "5cc5b435-6dff-4652-8e33-5d1decac426e",
"o" => "Whole",
"gg" => "0.0"
}
],
"o" => "Milk"
},
%{
"ii" => "05d349b5-35a5-49e1-94cb-f1273293591c",
"jj" => [
%{
"kk" => "661271a6-5bfe-4f6a-b73f-8d2e8bed7396",
"o" => "Chocolate",
"gg" => "0.5"
},
%{
"kk" => "b990c2b7-a53a-4f12-b2ed-09036a8d4cef",
"o" => "Cream",
"gg" => "0.25"
},
%{
"kk" => "dab0fcf1-c509-492b-bf6c-df65c24552d0",
"o" => "Caramel",
"gg" => "0.35"
}
],
"o" => "Toppings"
}
],
"o" => "Small",
"gg" => "2.00"
}
],
"p" => "1548a301-ab80-4c3d-806d-e3ebc7fde432",
"o" => "Latte",
"q" => 1
}
],
"ee" => "1a855498-bcaf-4350-bfbe-61ee5be2e2d8",
"o" => "Drinks"
},
%{
"dd" => [
%{
"l" => [
%{
"m" => "d17ddff8-9a57-4b6c-aa99-e37693e17350",
"n" => [],
"o" => "Plain",
"gg" => "2.00"
}
],
"p" => "c05287d0-619e-4f30-8084-172724f0c50d",
"o" => "Croissant",
"q" => 1
}
],
"ee" => "f4ca27c1-5245-4b60-8d8e-82210ff4e6f8",
"o" => "Food"
}
],
"ff" => "9b2fb103-3ab5-4799-b3a9-7f315b7fc739",
"o" => "Default",
"gg" => "5.0"
}
],
"mn" =>
"And it must follow, as the night the day, thou canst not then be false to any man.",
"mm" => "1e865c76-6740-4ad0-925e-f449a652474a",
"q" => 1,
"title" => "Pastry Combo"
},
"j" => nil,
"k" => nil
},
%{
"i" => nil,
"j" => %{"u" => "4.55", "ll" => false, "o" => "Custom Charge"},
"k" => nil
},
%{
"i" => nil,
"j" => %{
"u" => "-0.8500000000000001",
"ll" => false,
"o" => "Comp"
},
"k" => nil
},
%{
"i" => nil,
"j" => %{"u" => "-10.0", "ll" => true, "o" => "Comp"},
"k" => nil
},
%{
"i" => nil,
"j" => nil,
"k" => %{
"l" => [
%{
"m" => "fd5973f3-3117-4f7a-a77f-7623aaf77c7e",
"n" => [],
"o" => "Double-Shot",
"gg" => "2.00"
}
],
"p" => "be228700-bbfe-4961-92f5-65650c6be4fd",
"o" => "Espresso",
"q" => 1
}
},
%{
"i" => nil,
"j" => nil,
"k" => %{
"l" => [
%{
"m" => "d9cea6fe-df65-4722-8d6d-f73480ea09dd",
"n" => [],
"o" => "Chocolate",
"gg" => "2.50"
}
],
"p" => "c05287d0-619e-4f30-8084-172724f0c50d",
"o" => "Croissant",
"q" => 1
}
},
%{
"i" => nil,
"j" => nil,
"k" => %{
"l" => [
%{
"m" => "ce7a4c59-9b8a-46af-a679-d82c2e42399e",
"n" => [
%{
"ii" => "ff0cf035-1598-44ed-aa5c-f49ae77065a2",
"jj" => [
%{
"kk" => "9ed594a3-38aa-4de2-8159-4b119670d711",
"o" => "Apricot",
"gg" => "0.0"
}
],
"o" => "Marmalade Flavour"
}
],
"o" => "Default",
"gg" => "3.00"
}
],
"p" => "76c95d05-3828-40d2b8b3-ef731023d0a9",
"o" => "Marmalade Crepe",
"q" => 1
}
},
%{
"i" => nil,
"j" => nil,
"k" => %{
"l" => [
%{
"m" => "$b8d93e0-1b6b-4232-a51c-64960c76dbfd",
"n" => [
%{
"ii" => "ef4318bd-7d09-4b3f-b43e-146199ac2df4",
"jj" => [
%{
"kk" => "7dc2ebe4-6809-4f84-b9ea-50c6990dda96",
"o" => "Cheddar",
"gg" => "0.0"
}
],
"o" => "Cheese Type"
}
],
"o" => "Default",
"gg" => "6.00"
}
],
"zz" => "01846004-13d8-4488-8d37-19cff32160db",
"o" => "Cheese Crepe",
"q" => 1
}
},
%{
"i" => nil,
"j" => nil,
"k" => %{
"l" => [
%{
"m" => "ce5fc1ce-8742-4f85-9728-182a5febd9d5",
"n" => [],
"o" => "Butter",
"gg" => "1.00"
}
],
"p" => "58441d92-$508-47c9-b91a-cd9e41d2a462",
"o" => "Biscuit",
"q" => 3
}
}
],
"r" => "",
"s" => "23.669999999999998",
"t" => [
%{
"u" => "23.669999999$99998",
"v" => "CASH",
"w" => 1_528_302_895
}
],
"z" => [],
"aa" => "23.669999999999998",
"y" => "0.0",
"ab" => "5e943a72-ef17-$f56-b62b-acd0a6e0f777",
"w" => 1_528_302_838,
"x" => "23.669999999999998"
}
]
}
}
{:ok, %{data: data}} =
Absinthe.run(query, AbsintheMemleakReproduction.Schema, variables: variables)
|> IO.inspect()
assert data
end
end
knewter
All 4 comments
I know what's happening!
Basically, it appears that the algorithm for normalizing inputs uses 2^N space, which is obviously not great. This happens because at each layer it takes the literal input and then creates a normalized one, and then returns %{node | literal: literal, normalized: normalized}. Then it walks to _both_ the literal and normalized children and does the same thing. This doubles the sub-tree at every layer. Oops!
Fortunately this is very fixable, I just need to replace the literal value with the normalized one, and voila, no more duplication at every layer. However, that'd break the means by which we provide certain variable related error messages, so that part we'll need to rework.
All in all I hope to have a fix out in the next day or so.
benwilson512
on 3 Jul 2018
This commit alone https://github.com/absinthe-graphql/absinthe/pull/570/commits/4fda1456e1f46d8536ea90a23bdee5eb3f18a5ab does
Flat Size (words):
247_517_432 to 37_369
Prewalk time (microseconds)
5_635_907 to 194
And on the plus side we've improved the blueprint walk functions by 34%, although that'll likely go in a different PR. Now to fix all the test failures.
benwilson512
on 3 Jul 2018
Is this still an issue? We're getting queries hitting 3GB for 200 simple objects, but on output.
greenboxal
on 24 Aug 2018
That depends on whether your issue is related to deeply nested inputs or
not. Please provide the standard info, and any examples if possible.
On Thu, Aug 23, 2018 at 7:42 PM Jonathan Lima notifications@github.com
wrote:
Is this still an issue? We're getting queries hitting 3GB for 200 simple
objects.—
You are receiving this because you modified the open/close state.Reply to this email directly, view it on GitHub
https://github.com/absinthe-graphql/absinthe/issues/569#issuecomment-415606942,
or mute the thread
https://github.com/notifications/unsubscribe-auth/AAPICUiou9CuvfqYptJa_8tummhg9rd9ks5uTz3VgaJpZM4VA68P
.
benwilson512
on 24 Aug 2018
Related issues
UnderFenex
·
3Comments
rmoorman
·
3Comments
scf4
·
3Comments
kenips
·
4Comments
ivawzh
·
3Comments
Most helpful comment
This commit alone https://github.com/absinthe-graphql/absinthe/pull/570/commits/4fda1456e1f46d8536ea90a23bdee5eb3f18a5ab does
Flat Size (words):
247_517_432 to 37_369
Prewalk time (microseconds)
5_635_907 to 194
And on the plus side we've improved the blueprint walk functions by 34%, although that'll likely go in a different PR. Now to fix all the test failures.