It is crucial that VersionedCrdsValue::insert_timestamp does not go
backward in time:
https://github.com/solana-labs/solana/blob/ec37a843a/core/src/crds.rs#L67-L79
Otherwise methods such as get_votes and get_epoch_slots_since will
break, which will break their downstream flow, including vote-listener
and optimistic confirmation:
https://github.com/solana-labs/solana/blob/ec37a843a/core/src/cluster_info.rs#L1197-L1215https://github.com/solana-labs/solana/blob/ec37a843a/core/src/cluster_info.rs#L1274-L1298
For that, Crds::new_versioned is intended to be called "atomically" with
Crds::insert_verioned (as the comment already says so):
https://github.com/solana-labs/solana/blob/ec37a843a/core/src/crds.rs#L126-L129
However, currently this is violated in the code. For example,
filter_pull_responses creates VersionedCrdsValues (with the current
timestamp), then acquires an exclusive lock on gossip, then
process_pull_responses writes those values to the crds table:
https://github.com/solana-labs/solana/blob/ec37a843a/core/src/cluster_info.rs#L2375-L2392
Depending on the workload and lock contention, the insert_timestamps may
well be in the past when these values finally are inserted into gossip.
To avoid such scenarios, this commit:
* removes Crds::new_versioned and Crd::insert_versioned.
* makes VersionedCrdsValue constructor private, only invoked in
Crds::insert, so that insert_timestamp is populated right before
insert.
This will improve insert_timestamp monotonicity as long as Crds::insert
is not called with a stalled timestamp. Following commits may further
improve this by calling timestamp() inside Crds::insert, and/or
switching to std::time::Instant which guarantees monotonicity.
Local timestamps are updated for records associated with a pubkey if the
origin is still active:
https://github.com/solana-labs/solana/blob/c8ed14c64/core/src/crds.rs#L301-L311
However this is done inconsistently on some gossip paths (pull requests
and pull responses) but not all (e.g. push messages). Additionally
update_record_timestamp is inefficient since there can be ~800 values
associated with each pubkey.
This commit updates records timestamps only on contact-infos; and,
instead utilizes origin's timestamp when purging old values.
record_labels returns all the possible labels for a record identified by
a pubkey, used in updating timestamp of crds values:
https://github.com/solana-labs/solana/blob/1792100e2/core/src/crds_value.rs#L560-L577https://github.com/solana-labs/solana/blob/1792100e2/core/src/crds.rs#L240-L251
The code relies on CrdsValueLabel to be limited to a small deterministic
set of possible values for a fixed pubkey. As we expand crds values to
include duplicate shreds, this limits what the duplicate proofs can be
keyed by in the table.
In addition the computation of these labels is inefficient and will
become more so as duplicate shreds and more types of crds values are
added. An alternative is to maintain an index of all crds values
associated with a pubkey.
process_pull_requests acquires a write lock on crds table to update
records timestamp for each of the pull-request callers:
https://github.com/solana-labs/solana/blob/3087c9049/core/src/crds_gossip_pull.rs#L287-L300
However, pull-requests overlap a lot in callers and this function ends
up doing a lot of redundant duplicate work.
This commit obtains unique callers before acquiring an exclusive lock on
crds table.
In several places in gossip code, the entire crds table is scanned only
to filter out nodes' contact infos. Currently on mainnet, crds table is
of size ~70k, while there are only ~470 nodes. So the full table scan is
inefficient. Instead we may maintain an index of only nodes' contact
infos.
Current code only returns values which are expired based on the default
timeout. Example from the added unit test:
- value inserted at time 0
- pubkey specific timeout = 1
- default timeout = 3
Then at now = 2, the value is expired, but the function fails to return
the value because it compares with the default timeout.
filter_crds_values checks every crds filter against every hash value:
https://github.com/solana-labs/solana/blob/ee646aa7/core/src/crds_gossip_pull.rs#L432
which can be inefficient if the filter's bit-mask only matches small
portion of the entire crds table.
This commit shards crds values into separate tables based on shard_bits
first bits of their hash prefix. Given a (mask, mask_bits) filter,
filtering crds can be done by inspecting only relevant shards.
If CrdsFilter.mask_bits <= shard_bits, then precisely only the crds
values which match (mask, mask_bits) bit pattern are traversed.
If CrdsFilter.mask_bits > shard_bits, then approximately only
1/2^shard_bits of crds values are inspected.
Benchmarking on a gce cluster of 20 nodes, I see ~10% improvement in
generate_pull_responses metric, but with larger clusters, crds table and
2^mask_bits are both larger, so the impact should be more significant.
* Gossip benchmark
* Rayon tweaking
* push pulls
* fanout to max nodes
* fixup! fanout to max nodes
* fixup! fixup! fanout to max nodes
* update
* multi vote test
* fixup prune
* fast propagation
* fixups
* compute up to 95%
* test for specific tx
* stats
* stats
* fixed tests
* rename
* track a lagging view of which nodes have the local node in their active set in the local received_cache
* test fixups
* dups are old now
* dont prune your own origin
* send vote to tpu
* tests
* fixed tests
* fixed test
* update
* ignore scale
* lint
* fixup
* fixup
* fixup
* cleanup
Co-authored-by: Stephen Akridge <sakridge@gmail.com>
* Batch process pull responses
* Generate pull requests at 1/2 rate
* Do filtering work of process_pull_response in read lock
Only take write lock to insert if needed.
