experiment: perf and allocation counter based idl decoder limiting #4624
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
crusso
commented
Jul 22, 2024
src/codegen/compile.ml
Outdated
| @@ -5131,7 +5131,7 @@ module IC = struct | |||
| ) | |||
|
|
|||
| (* For debugging *) | |||
| let _compile_static_print env s = | |||
| let compile_static_print env s = | |||
…ug); modify skip_any_vec to do limit checks
crusso
commented
Aug 2, 2024
crusso
changed the title
crusso
commented
Aug 8, 2024
| @@ -314,6 +319,9 @@ unsafe fn skip_any_vec(buf: *mut Buf, typtbl: *mut *mut u8, t: i32, count: u32) | |||
| // makes no progress. No point in calling it over and over again. | |||
| // (This is easier to detect this way than by analyzing the type table, | |||
| // where we’d have to chase single-field-records.) | |||
| for _ in 1..count { | |||
There was a problem hiding this comment.
Better might be to measure the cost of the previous skip and lower the limits based on a multiplier of that cost and return (without this faux for_loop). Then we would exit even earlier!
4 tasks
mergify bot
pushed a commit
that referenced
this pull request
Aug 16, 2024
Simplifies #4624 to a simple linear limit on the number of decoded values as a function of decoded payload size, instead of using two linear functions on perfcounter (simulated or real) and allocation counter. The function is: value_quota(blob) : Nat64 = blob.size() * (numerator/denominator) + bias where blob is the candid blob to be decoded, and `numerator` (default 1), `denominator` (default 1) and `bias` (default 1024) are `Nat32s`. Much simpler than #4624 and doesn't depend on vagaries of instruction metering and byte allocation which varies with gc and compiler options, but is it good enough? The constants can be (globally) modified/inspected using prims (Prim.getCandidLimits/Prim.setCandidLimits) which will need to get exposed in base eventually. The quota is decremented on every call to deserialise or skip a value in vanilla candid mode (destabilization is not metered). The quota is eagerly checked before deserializing or skipping arrays. One possible refinement would be to combine the value quota with a memory quota (though the latter would still vary with gc flavour and perhaps word-size unless we count logical words) - [x] Disable for destabilization (iff Registers.get_rel_buf_opt is zero) - [x] Port new candid spacebomb test suite to drun-tests, to test against real perf counter provided by drun. - [x] Bump candid dependency to most recent - [x] Pass new spacebomb tests, both in candid test suite on wasmtime using value counter.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
periodically check the instruction counter during IDL decoding to prevent abuse