Downtime detector module #3603
Description
Background
For several use cases, we need a module that can detect when the chain is recovering from downtime. We want to be able to efficiently know "Has it been $RECOVERY_PERIOD minutes since the chain has been down for $DOWNTIME_PERIOD", and expose this as a query to contracts.
So for instance, you'd want to know if it has been at least 10 minutes, since the chain was down for > 30 minutes. Since you assume in such an event that it may take ~10 minutes for price oracles to be arb'd to correct.
Suggested Design
Theres a couple designs, such as:
- Iterating over block times from the last N blocks (with a heuristic filter based on average block time)
- Implies bounds on recovery time
- Linear iteration if heuristic is met
- Requires encoding expected block time
- Restricting downtime period, and storing a state entry for last time a downtime of length $D occurred
Because this may be in the hot-loop of contracts, I think we need to go with the approach that has minimal query compute, which is the latter. So I'll explain that in more depth.
We restrict the $DOWNTIME_PERIOD options that you can query, to be say 30seconds, 1 min, 2 min, 3 min, 4 min, 5 min, 10 min, 20 min, 30 min, 40 min, 50 min, 1 hr, 1.5hr, 2 hr, 2.5 hr, 3 hr, 4 hr, 5 hr, 6 hr, 9hr, 12hr, 18hr, 24hr, 36hr, 48hr
In the downtime detector module, we store state entries for:
- Last blocks timestamp
- For each period, last time there was downtime
Then in every begin block:
- Store last blocks timestamp
- if time since last block timestamp >= 30 seconds, iterate through all $DOWNTIME_PERIODS and add a state entry for the current block time
Then our query for has it been $RECOVERY_PERIOD since $DOWNTIME_PERIOD, simply reads the state entry for that $DOWNTIME_PERIOD, and then checks if time difference between now and that block is > RECOVERY_PERIOD.
Acceptance Criteria
- Downtime detector module exists
- Has query for
RECOVERY_SINCE_DOWNTIME_LENGTH - Query is efficient