anmolnar commented on pull request #4125: URL: https://github.com/apache/hbase/pull/4125#issuecomment-1074504042
Okay, another round of investigation finished. I think I pinned down the issue: So, the problem is not the timing of the Netty pipeline become usable, because Netty already takes care of the write requests from this aspect. The root problem is that we're trying to send a Call, not a ByteBuf which has to be converted, so either the `BufferCallBeforeInitHandler` or the protobuf converter has to be on the pipeline before calling the write (and flush). This was the case previously, because a single channel handler (BufferCallBeforeInitHandler) was added in the connect() method which means it's immediately added to the pipeline. Now we need the SSL handler as well and this must be done via `ChannelInitializer` (HBaseClientPipelineFactory). As a consequence the channel pipeline will be built fully asynchronously by Netty and connect() method will only return a promise without building anything. We'll add the call to the "builder" pipeline which has the "head", "the pipeline factory" and the "tail". None of them takes care of the Call and fails. How to synchronize this? I see some options: 1. Using the `sync()` method to force building the usable pipeline first. 2. Add latch - ugly and same effect 3. Try to upgrade Netty - HBase uses 4.0 which is quite old, maybe they're doing this better in a later version. I'll check. 4. Put the `writeAndFlush()` operation in a channel promise callback to make sure it's only called when the pipeline is ready to use. This seems to me the most Netty-way resolution. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: issues-unsubscr...@hbase.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org
