feat: remove TEE from tree #2627
Conversation
TEE input producer jobs are now created by monitoring the db instead of hooking directly into the tree logic.
Change get_new_l1_batches() to only return a single value.
thomasknauth
force-pushed
the
tk-remove-tee-from-tree
branch
from
3a1fa96 to
c1988f7
Compare
| // Since we depend on Merkle paths, we use the proof_generation_details table to inform us of newly available to-be-proven batches. | ||
| let row = sqlx::query!( | ||
| r#" | ||
| SELECT |
There was a problem hiding this comment.
Wouldn't this mean you could miss some entries (you query once, MAX = 10, you query second time, now there have been 5 batches incoming, MAX = 15, you'd be missing batches 11, 12, 13 and 14, no?).
There was a problem hiding this comment.
That should not happen. You only return max(l1_batch_number) from the proof_generation_details table if the TEE jobs table is empty (which just handles the special case when the thing comes up for the first time, L114). Normal case is to return max(l1_batch_number)+1 from the TEE jobs table.
| SELECT | ||
| MAX(l1_batch_number) AS "number" | ||
| FROM | ||
| proof_generation_details |
There was a problem hiding this comment.
I don't like that TEE is tied to proof_generation_details. If this table happens to become too hot (for instance, a bug in TEE), proof generation (and our SLAs + customer experience) will degrade. I don't have a better solution than code duplication, which I dislike even more.
There was a problem hiding this comment.
What code do you intend to duplicate as a potential solution? I'm open to suggestions in terms of an alternate "signal" to monitor for new batches.
There was a problem hiding this comment.
I think that the right way would be to move save_merkle_paths_artifacts_metadata from proof_generation_dal to the blocks_dal (e.g. existence of the metainformation is the property of the block, regardless of the consumer).
This way we (later, not now!) will be able to get rid of insert_proof_generation_details in the updater.rs as well -- so that the tree is not aware of any verification logic (as it should be).
There was a problem hiding this comment.
I'd say it's unclear who would create the tables afterwards. But I don't oppose the suggestion.
TEE input producer jobs are now created by monitoring the db instead of hooking directly into the tree logic.
What ❔
Why ❔
Checklist
zk fmtandzk lint.