Garbage-collect batch DB #502
Conversation
Signed-off-by: Matej Pavlovic <matopavlovic@gmail.com>
Signed-off-by: Matej Pavlovic <matopavlovic@gmail.com>
Signed-off-by: Matej Pavlovic <matopavlovic@gmail.com>
Signed-off-by: Matej Pavlovic <matopavlovic@gmail.com>
Signed-off-by: Matej Pavlovic <matopavlovic@gmail.com>
Signed-off-by: Matej Pavlovic <matopavlovic@gmail.com>
Signed-off-by: Matej Pavlovic <matopavlovic@gmail.com>
Signed-off-by: Matej Pavlovic <matopavlovic@gmail.com>
Signed-off-by: Matej Pavlovic <matopavlovic@gmail.com>
| state.batchStore[batchID] = &b | ||
| state.batchesByRetIdx[retIdx] = append(state.batchesByRetIdx[retIdx], batchID) | ||
| } | ||
|
|
||
| batchdbpbdsl.BatchStored(m, origin.Module, origin) |
There was a problem hiding this comment.
Seems weird that if too old we do not store it but we say we have stored it. Have you considered whether this could break somewhere else? Probably not because if then it is not stored we just ask around for it. However, intuitively, this should never happen?
There was a problem hiding this comment.
Oh yes, very good point, I was actually thinking about it the very same way and wanted to write an explanatory comment in the code, but I forgot do do that.
I agree that it's a bit counter-intuitive. The thing is that if the BatchDB's retention index exceeded the retention index of the batch, it's equivalent to the batch having been stored and already garbage-collected. In fact, this might even really happen, when the batch is concurrently being stored by the availability module's dissemination and retrieval functions (the dissemination to this particular node only needs to be slightly late). Both of these functions need to believe to have stored the batch in order to properly continue (although it's likely that nobody cares about their output any more, as they are being garbage-collected too).
There was a problem hiding this comment.
I added the explanatory comment, thanks for pointing it out.
Signed-off-by: Matej Pavlovic <matopavlovic@gmail.com>
The Batch DB also needs to be garbage-collected.
The current implementation stores all batches ever delivered in memory and thus runs out of memory rather quickly under high load.