Client/Ingest node buffer/backpressure handling



If we have a separate data node cluster with dedicated ingest and dedicated client node, in the event that the data node cluster becomes unavailable (for whatever reason), what will happen to the documents which have been passed over to the client/ingest node?

Will they simply be dropped or will it try to buffer for a certain limit?


(Martijn Van Groningen) #2

The index / bulk requests in flight between the client and the data node will eventually fail and return with an error to the client. When the data node becomes unavailable the requests in flight will not fail immediately but ES will wait for upto 60 seconds for the data node to reappear. If that doesn't happen the request will be failed and otherwise the index requests will eventually be processed.

From the outside it doesn't really matter if there is an ingest node in the picture. The requests are then temporarily being kept in the ingest node instead of the client node.

(system) #3

This topic was automatically closed 28 days after the last reply. New replies are no longer allowed.