Serge/liquidator split tcs and liquidation #914
Conversation
github-actions
bot
added
dependency
farnyser
force-pushed
the
serge/liquidator-split-tcs-and-liquidation
branch
from
d19ccd3 to
3894d8f
Compare
farnyser
force-pushed
the
serge/liquidator-split-tcs-and-liquidation
branch
from
3894d8f to
020252b
Compare
| // a task must be available to process | ||
| // find it in global shared state, and mark it as processing | ||
| let task = worker_pull_task(&shared_state, id, tcs_capable_workers, only_liquidation) | ||
| .expect("Worker woke up but has nothing to do"); |
There was a problem hiding this comment.
I think with the tcs batching pulled back in, this no longer works. You send a trigger for each tcs candidate, but the processing tasks eat up multiple at once - so for N candidates the there'll be fewer than N worker tasks.
There's also some weirdness now where the tcs jobs trigger on the tcs channel and take all the liquidation jobs and then the liquidation job wakes up (on the liq channel) and has no work to do.
It may be more robust to avoid the requirement that the number of messages on the channels and the number of work chunks match up exactly.
There was a problem hiding this comment.
It "works" because the TCS task will just be made of an empty vector (no candidate) if everything was consumed before
There was a problem hiding this comment.
Ah, I see! Would you be up for making it more explicit by adding a no-work WorkerTask?
| // a task must be available to process | ||
| // find it in global shared state, and mark it as processing | ||
| let task = worker_pull_task(&shared_state, id, tcs_capable_workers, only_liquidation) | ||
| .expect("Worker woke up but has nothing to do"); |
There was a problem hiding this comment.
Ah, I see! Would you be up for making it more explicit by adding a no-work WorkerTask?
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)
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)
No description provided.