Closed shreyan-gupta closed 2 days ago
All modified and coverable lines are covered by tests :white_check_mark:
Project coverage is 71.44%. Comparing base (
fecce01
) to head (567d45f
). Report is 16 commits behind head on master.
:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.
🚨 Try these New Features:
In future PRs we plan to get rid of the caches in ChainStore as we don't believe them to be too useful and is a complication to the current codebase.
As a pre-step, this PR gets rid of the
no_cache
feature. This was being used in ChainStore TrieStore, and EpochManager.Here are the associated caches that are affected
Effect on Parameter Estimator
While the param estimator hasn't been run for a long time since the introduction of memtries, it used to rely on the
no_cache
feature to effectively test the worst case performance with the cache removed from Trie.With the introduction on memtrie, this check is now redundant as we no longer use the Trie Store Cache after the introduction of memtrie.
However, as of today, some nodes like RPC nodes (which track all shards) do not use memtries and this might be relevant for them. That said, it's probably not worth it to block this change on potential future use of
no_cache
for RPC nodes where we are moving towards a memtrie world.Native transfer workloads
This change is backed up by Moritz's tests on native transfer workload with 10k accounts and unlimited configs.
This shows that the caches don't have a significant impact on the workloads.