Known issue: orphan state witness #15
Labels
Comments
github-merge-queue bot
pushed a commit
to near/nearcore
that referenced
this issue
Feb 23, 2024
### Description This PR adds a pool for orphaned `ChunkStateWitnesses`. To process a `ChunkStateWitness` we need the previous block, but sometimes it isn't available immediately. The node might receive a `ChunkStateWitness` before the block that's required to process it. In such cases the witness becomes an "orphaned chunk state witness" and it's put in `OrphanChunkStateWitnessPool`, where it waits for the desired block to appear. Once a new block is accepted, we fetch all orphaned witnesses that were waiting for this block from the pool and process them. ### Design of `OrphanStateWitnessPool` `OrphanStateWitnessPool` keeps a cache which maps `shard_id` and `height` to an orphan `ChunkStateWitness` with these parameters: ```rust witness_cache: LruCache<(ShardId, BlockHeight), ChunkStateWitness>, ``` All `ChunkStateWitnesses` go through basic validation before being put in the orphan cache. * The signature is checked to make sure that this witness really comes from the right chunk producer that should produce a witness at this height and shard_id. * Client keeps only witnesses which are within 5 blocks of the current chain head to prevent spam attacks. Without this limitation a single malicious chunk producer could fill the whole cache with their fake witnesses. * There's also a limitation on witness size to limit the amount of memory consumed by the pool. During StatelessNet loadtests performed by `@staffik` and `@Longarithm` the observed `ChunkStateWitness` sIze was 16-32MB, so a 40MB limit should be alright. This PR only limits the size of orphaned witnesses, limiting the size of non-orphan witnesses is much more tricky, see the discussion in #10615. It's impossible to fully validate an orphaned witness, but this partial validation should be enough to protect against attacks on the orphan pool. Under normal circumstances there should be only a few orphaned witnesses per shard. If the node has fallen behind by more than a few blocks, it has to catch up and its chunk endorsements don't matter. The default cache capacity is set to 25 witnesses. With 5 shards it provides capacity for 5 orphaned witnesses on each shard, which should be enough. Assuming that a single witness can take up 40 MB, the pool will consume at most 1GB at full capacity. The changes are divided into individual commits, they can be reviewed commit-by-commit. ### Fixes Fixes: #10552 Fixes: near/stakewars-iv#15
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Bug Report
Overview
Currently we don't properly handle the case when state witness arrives before the parent block.
For more information see near/nearcore#10552
Impact
This results in chunk validator not sending endorsement to block producer which could cause chunk not to be included in the block.
The text was updated successfully, but these errors were encountered: