Some systems have a very small /dev/shm, for example, see:
https://github.com/docker-library/postgres/issues/416
So we should just use the temporary directory on all operating systems.
Also:
* use TempDir to generate the temporary path
* delete the code that we copied from sled
* prefix the temporary path with the state version and network
## Motivation
Prior to this PR we've been using `sled` as our database for storing persistent chain data on the disk between boots. We picked sled over rocksdb to minimize our c++ dependencies despite it being a less mature codebase. The theory was if it worked well enough we'd prefer to have a pure rust codebase, but if we ever ran into problems we knew we could easily swap it out with rocksdb.
Well, we ran into problems. Sled's memory usage was particularly high, and it seemed to be leaking memory. On top of all that, the performance for writes was pretty poor, causing us to become bottle-necked on sled instead of the network.
## Solution
This PR replaces `sled` with `rocksdb`. We've seen a 10x improvement in memory usage out of the box, no more leaking, and much better write performance. With this change writing chain data to disk is no longer a limiting factor in how quickly we can sync the chain.
The code in this pull request has:
- [x] Documentation Comments
- [x] Unit Tests and Property Tests
## Review
@hdevalence
This change explicitly documents cancellation contracts for our Tower services,
and tries to correct a bug in the implementation of the CheckpointVerifier,
which duplicates information from the state service but did not ensure that it
would be kept in sync.
This change has two benefits:
* reduces conflicts with the sled refactor and any replacement
* allows the function to be called independently for testing
`check_contextual_validity` mistakenly used the new block's hash to try
to get the parent block from the state. This caused a panic, because the
new block isn't in the state yet.
Use `StateService::chain` to get the parent block, because we'll be
using `chain` for difficulty adjustment contextual verification anyway.
* Add internal iterator API for accessing relevant chain blocks
* get blocks from all chains in non_finalized state
* Impl FusedIterator for service::Iter
* impl ExactSizedIterator for service::Iter
* let size_hint find heights in side chains
Co-authored-by: teor <teor@riseup.net>
* Add transcript test for requests while state is empty
* Add happy path test for each query once the state is populated
* let populate logic handle out of order blocks
* Add a maximum queued height metric to the finalized state
And rename all the finalized state metrics to contain "finalized".
* Use i32 and -1 instead of Option<Height>
Co-authored-by: Jane Lusby <jlusby42@gmail.com>
Prior to this PR we realized that the RFC had been drafted with the assumption that chains would be ordered from best to worst in `NonFinalizedState`. This assumption was incorrect, since `BTreeSet` only ever orders values in ascending order. This discrepancy was noticed and fixed in the code, but there were still some inconsistencies that needed to be cleaned up.
This PR updates all the incorrect or confusing comments about chain ordering in the RFC and code.
Prior to this PR `memory_state` defined and implemented functionality for three different types, `Chain`, `NonFinalizedState`, and `QueuedBlocks`. Each of these components will need a fair number of unit tests, and I realized that as its currently organized it would be difficult to organize the tests or at a glance figure out which tests are testing which components.
This PR changes the organization of `memory_state` such that each component it exports is defined in its own module. In follow up PRs each module will get its own test module, which will focus exclusively on unit tests for the item defined there-in.
- [Tracking Issue](https://github.com/ZcashFoundation/zebra/issues/1250)