Use vertx blockingHandlers to run Bookkeeper http handlers which could be blocking

[lhotari]

Motivation

• The http handler implementations in Bookkeeper aren't necessarily non-blocking. That's why they should be executed on the blocking thread pool in vertx.

Changes

Executing on blocking thread pool in vertx can be achieved by registering the handlers using the blockingHandler method instead of the handler method.

[dragonls]

In our case, we are using /api/v1/ledger/list to get the ledgers in bookie, which may run more than 30 seconds due to lots of ledgers. At the same time, all the other http request will be blocked, such as /metrics or even /heartbeat. It seems that all http requests in bookie is processed by a single thread.

We found that if we update the code to:

        HttpRouter<VertxAbstractHandler> requestRouter = new HttpRouter<VertxAbstractHandler>(handlerFactory) {
            @Override
            public void bindHandler(String endpoint, VertxAbstractHandler handler) {
                router.get(endpoint).blockingHandler(handler, false);
                router.put(endpoint).blockingHandler(handler, false);
                router.post(endpoint).blockingHandler(handler, false);
                router.delete(endpoint).blockingHandler(handler, false);
            }
        };

then the http requests could be run concurrently by updating the second parameter ordered from true to false.

I am not sure why the bookie process all the http request by a single thread. Is it necessarily?