Commit Graph

10 Commits

Author SHA1 Message Date
Luke Parker
e772b8a5f7 #560 take two, now that #560 has been reverted (#561)
* Clear upons upon round, not block

* Cache the proposal for a round

* Rebase onto develop, which reverted this PR, and re-apply this PR

* Set participation upon participation instead of constantly recalculating

* Cache message instances

* Add missing txn commit

Identified by @akildemir.

* Correct clippy lint identified upon rebase

* Fix tendermint chain sync (#581)

* fix p2p Reqres protocol

* stabilize tributary chain sync

* fix pr comments

---------

Co-authored-by: akildemir <34187742+akildemir@users.noreply.github.com>
2024-07-16 19:42:15 -04:00
Luke Parker
c706d8664a Use OptimisticTransactionDb
Exposes flush calls.

Adds safety, at the cost of a panic risk, as multiple TXNs simultaneously
writing to a key will now cause a panic. This should be fine and the safety is
appreciated.
2024-03-20 23:42:40 -04:00
Luke Parker
157acc47ca More aggresive WAL parameters 2024-03-09 05:05:43 -05:00
Luke Parker
68dc872b88 sync every txn 2024-03-09 03:18:52 -05:00
Luke Parker
97f433c694 Redo how WAL/logs are limited by the DB
Adds a patch to the latest rocksdb.
2024-03-09 02:20:14 -05:00
Luke Parker
10f5ec51ca Explicitly limit RocksDB logs 2024-03-08 09:19:34 -05:00
Luke Parker
6d5851a9ee Use lz4 instead of zstd for the DB
zstd was recommended for the base layer only, due to its CPU requirements. That
was a misreading on mhy behalf.

lz4 gets ~5% better compression than snappy with ~30% faster performance. zstd
does ~25% better than lz4 yet at ~30% of the performance.
2023-07-26 14:05:10 -04:00
Luke Parker
821f5d8de4 Restore create_if_missing to RocksDB code 2023-07-25 23:00:10 -04:00
Luke Parker
32435d8a4c Consolidate RockDB code
Moves explicitly to zstd. RockDB recommends zstd, or at least lz4 over snappy,
and this minimizes which dependencies we pull in.
2023-07-25 21:43:27 -04:00
Luke Parker
62504b2622 Bind RocksDB into serai-db 2023-07-13 19:09:11 -04:00