liquidator: split TCS triggering and liquidation job
Concurrent execution of candidate lookup and tx building/sending
- Also added an health assertion IX to protect liqor in multi liquidation scenario
- And a timeout for jupiter v6 queries (avoid blocking liquidation because of slow TCS)
* rename usd_opt to usdc_opt in OracleAccountInfos
* use fallbacks when fetching bank+ price in AccountFetcher struct
* feat: add derive_fallback_oracle_keys to MangoGroupContext
* test: properly assert failure CU in test_health_compute_tokens_fallback_oracles
* provide fallback oracle accounts in the rust client
* liquidator: update for fallback oracles
* set fallback config in mango services
* support fallback oracles in rust settler + keeper
* fix send error related to fetching fallbacks dynamically in tcs_start
* drop derive_fallback_oracle_keys_sync
* add fetch_multiple_accounts to AccountFetcher trait
* revert client::new() api
* deriving oracle keys uses account_fetcher
* use client helpers for deriving health_check account_metas
* add health_cache helper to mango client
* add get_slot to account_fetcher
* lint
* cached account fetcher only fetches uncached accounts
* ensure keeper client does not use cached oracles for staleness checks
* address minor review comments
* create unique job keys for CachedAccountFetcher.fetch_multiple_accounts
* fmt
* improve hashing in CachedAccountFetcher.fetch_multiple_accounts
---------
Co-authored-by: Christian Kamm <mail@ckamm.de>
- Don't just restart on new listings, but also on significant changes to
old listings such as oracle changes.
- Cover the liquidator and settler in addition to the keeper.
(cherry picked from commit ce16d79b13)
Many instructions now return PreparedInstructions instead of a direct
Instruction or Vec<Instruction>. That way they can keep track of the
expected cu cost of the instructions for the compute limit instruction
that gets added once all instructions are made.
Serge Farny
Christian Kamm
Lou-Kamades