Add a test for the mempool dynamic memory usage
AbandonedPublic
Actions

Authored by Fabien on May 7 2020, 18:37.

Details

Reviewers

deadalnix

Group Reviewers

Restricted Project

Summary

The mempool max size is determined by evaluating the mempool dynamic memory usage. This cause the behavior to change between various architectures that have different pointer sizes or different internal representations of the standard library structures. This diff adds a test for the mempool dynamic memory usage calculation, as a first step towards faking this calculation to make it architecture independent.

This test is only valid for glibc based 64 bits architectures for now, and skipped for others.
The goal is to move the mempool memory accounting in such a way that the test will pass on any target.

Test Plan

ninja check-bitcoin-mempool_tests

Diff Detail

Repository

rABC Bitcoin ABC

Branch

add_mempool_dynamic_usage_tests

Lint

Lint Passed

Unit

No Test Coverage

Build Status

Buildable 10641
Build 19089: Default Diff Build & Tests
Build 19088: arc lint + arc unit

Event Timeline

Fabien created this revision.May 7 2020, 18:37

Herald added a reviewer: Restricted Project. · View Herald TranscriptMay 7 2020, 18:37

Fabien requested review of this revision.May 7 2020, 18:37

Build Bitcoin-ABC / Diffs / Diff Testing (build-diff) started.

Build Bitcoin-ABC / Diffs / Diff Testing (build-diff) passed.

Harbormaster completed remote builds in B10641: Diff 19808.May 7 2020, 18:56

Why is that a problem that the memory usage is computed in an architecture dependent manner? Isn't it architecture dependent?

This revision now requires changes to proceed.May 9 2020, 11:50

The mempool dynamic memory usage is what is used to limit the number of transactions in the mempool (-maxmempool). This creates a disparity between 32 and 64 bits architectures because the memory usage is obviously architecture dependent.
I am trying to solve the issue you described in D5852, and in order to not break the API (-maxmempool) I need to fake the way the memory usage is evaluated by the mempool. This test is here to guarantee that the evaluation remains the same, so there is no change in behavior for the reference linux 64 arch, then after a few diffs it will demonstrate that the evaluation is now the same for all archs.
I updated the summary to make the intent more clear.