-
Notifications
You must be signed in to change notification settings - Fork 4.4k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Add log subscriptions to historian for thick clients #27
Comments
Since this ticket was opened, the historian's channel is now held by let first_id = stub.get_id(is_first=true);
loop {
stub.get_entries(id=first_id);
} We need to switch to a broadcast-oriented API where clients can subscribe for parts of the ledger, allowing the leader to control how much outbound traffic it is putting on the wire. Verifiers, on the other hand, may be less constrained and could offer a more flexible API. |
An incremental improvement would be to port the clients to tokio, so that our synchronous API can be implemented with futures instead of blocking calls. Futures are still clunky, however. What we really want is to leave our synchronous API as-is and use coroutines to interleave event handling. Unfortunately, it seems that's not too far along in Rust: https://github.com/rust-lang/rfcs/blob/master/text/2033-experimental-coroutines.md |
Another option is to switch the API to something closer to a one-way street. Clients would listen for ledger updates. The server could cycle over the full ledger continuously so that new clients can clone it, and existing clients could recover after a restart. If we did that, client code can stick to its synchronous API and block all it wants on the client side. |
@aeyakovenko, did you want to try piping the thin client API through the thick client mechanism in this upcoming release? I'm okay pushing that back, but if you think it'd be easy, I can give it a go. I'm reworking the thin client right now anyway. |
Fixed in edf6272 |
* Remove untested --no-wait feature * Make --transactions-db an option, not an arg So that in the future, we can make it optional * Remove more untested features Too many false positives in that santity check. Use --dry-run instead. * Add dry-run mode to ThinClient * Cleaner dry-run * Make key parameters required Just don't use them in --dry-run * Add option to write the transaction log --dry-run doesn't write to the database. Use this option if you want a copy of the transaction log before the final run. * Revert --transaction-log addition Implement solana-labs#27 first * Fix CI * Update readme * Fix CI in copypasta
* Initial commit * Execute transfers * Refactor for testing * Cleanup readme * Rewrite * Cleanup * Cleanup * Cleanup client * Use a Null Client to move prints closer to where messages are sent * Upgrade Solana * Move core functionality into its own module * Handle transaction errors * Merge allocations * Fixes * Cleanup readme * Fix markdown * Add example input * Add integration test - currently fails * Add integration test * Add metrics * Use RpcClient in dry-run, just don't send messages * More metrics * Fix dry run with no keys * Only require one approval if fee-payer is the sender keypair * Fix bugs * Don't create the transaction log if nothing to put into it; otherwise the next innvocation won't add the header * Apply previous transactions to allocations with matching recipients * Bail out of any account already has a balance * Polish * Add new 'balances' command * 9 decimal places * Add missing file * Better dry-run; keypair options now optional * Change field name from 'bid' to 'accepted' Also, tolerate precision change from 2 decimal places to 4 * Write to transaction log immediately * Rename allocations_csv to bids_csv So that we can bypass bids_csv with an allocations CSV file * Upgrade Solana * Remove faucet from integration test * Cleaner integration test Won't work until this lands and is released: #9717 * Update README * Add TravicCI script to build and test (#1) * Add distribute-stake command (#2) * Distribute -> DistributeTokens (#3) * Cache cargo deps (#4) * Add docs (#5) * Switch to latest Solana 1.1 release (#7) * distribute -> distribute-tokens (#9) * Switch from CSV to a pickledb database (#8) * Switch from CSV to a pickledb database * Allow PickleDb errors to bubble up * Dedup * Hoist db * Add finalized field to TransactionInfo * Don't allow RPC client to resign transactions * Remove dead code * Use transport::Result * Record unconfirmed transaction * Fix: separate stake account per allocation * Catch transport errors * Panic if we attempt to replay a transaction that hasn't been finalized * Attempt to fix CI PickleDb isn't calling flush() or close() after writing to files. No issue on MacOS, but looks racy in CI. * Revert "Attempt to fix CI" This reverts commit 1632394f636c54402b3578120e8817dd1660e19b. * Poll for signature before returning * Add --sol-for-fees option for stake distributions * Add --allocations-csv option (#14) * Add allocations-csv option * Add tests or GTFO * Apply review feedback * apply feedback * Add read_allocations function * Update arg_parser.rs * Fix balances command (#17) * Fix balances command * Fix readme * Add --force to transfer to non-empty accounts (#18) * Add --no-wait (#16) * Add ThinClient methods to implement --no-wait * Plumb --no-wait through No tests yet * Check transaction status on startup * Easier to test * Wait until transaction is finalized before checking if it failed with an error It's possible that a minority fork thinks it failed. * Add unit tests * Remove dead code and rustfmt * Don't flush database to file if doing a dry-run * Continue when transactions not yet finalized (#20) If those transactions are dropped, the next run will execute them. * Return the number of confirmations (#21) * Add read_allocations() unit-test (#22) Delete the copy-pasted top-level test. Fixes #19 * Add a CSV printer (#23) * Remove all the copypasta (#24) * Move resolve_distribute_stake_args into its own function * Add stake args to token args * Unify option names * Move Command::DistributeStake into DistributeTokens * Remove process_distribute_stake * Only unique signers * Use sender keypair to fund new fee-payer accounts * Unify distribute_tokens and distribute_stake * Rename print-database command to transaction-log (#25) * Send all transactions as quickly as possible, then wait (#26) * Send all transactions as quickly as possible, then wait * Exit when finalized or blockhashes have expired * Don't need blockhash in the CSV output * Better types CSV library was choking on Pubkey as a type. PickleDb doesn't have that problem. * Resend if blockhash has not expired * Attempt to fix CI * Move log to stderr * Add constructor, tuck away client (#30) * Add constructor, tuck away client * Fix unwrap() caught by CI * Fix optional option flagged as required * Bunch of cleanup (#31) * Remove untested --no-wait feature * Make --transactions-db an option, not an arg So that in the future, we can make it optional * Remove more untested features Too many false positives in that santity check. Use --dry-run instead. * Add dry-run mode to ThinClient * Cleaner dry-run * Make key parameters required Just don't use them in --dry-run * Add option to write the transaction log --dry-run doesn't write to the database. Use this option if you want a copy of the transaction log before the final run. * Revert --transaction-log addition Implement #27 first * Fix CI * Update readme * Fix CI in copypasta * Sort transaction log by finalized date (#33) * Make --transaction-db option implicit (#34) * Move db functionality into its own module (#35) * Move db functionality into its own module * Rename tokens module to commands * Version bump * Upgrade Solana * Add solana-tokens to build * Remove Cargo.lock * Remove vscode file * Remove TravisCI build script * Install solana-tokens Co-authored-by: Dan Albert <dan@solana.com>
* Initial memo commit * Add program id * Update CI * Remove unneeded dependencies * Try instead of map * clippy --tests
* Update SECURITY.md with anza links and names * Update obsolete section about shipping patches * Switch email to security@solana.com
Think clients and verifiers need access to the full log. Currently, the accountant receives new log entries by listening to a sync_channel held by the historian. Those entries are not serialized and do not cross the network.
Implementing this ticket might be accomplished by adding
historian_stub
and correspondinghistorian_skel
that listens on the internal channel. It's also possibly we might want a generic "switchboard operator" that simply maps network protocols to our deserialized, well-typed channels.The text was updated successfully, but these errors were encountered: