Explanation of Typesense's server parameters `--healthy-read-lag` and `--healthy-write-lag`.

To answer your question, when writes are sent to Typesense they are replicated to other nodes via Raft operations... So the writes are first written to an append-only log and then get applied to the in-memory index. If a large volume of writes show up, there might be a lag between writes being written to the log and when they actually get applied to memory. This lag is what is defined by those two parameters.

So if the lag is over --healthy-write-lag, then additional writes are rejected with an HTTP 503 while writes already in the log are applied, and lag eventually reduces below this threshold. If the lag is over --healthy-read-lag, then all reads (eg: searches) also start returning a 503. The former is useful if you don't want to overwhelm the cluster with writes and you want a backoff indicator for the system that is writing into Typesense. The latter is useful in cases where you'd rather not serve any stale data for searches and want to wait for the data that has been sent to Typesense to be available for search