java - storm topology: one to many (random) -
i'm using kafkaspout spout read (6) partitions on kafka topic. first bolt in topology has convert byte stream struct (via idl definition), lookup value in db , pass these values second bolt writes cassandra.
there several issues occurring:
- many fail(s) kafka spout.
- the first bolt reports "capacity" of > 2.0 storm ui.
i've tried increase parallelism appears storm accept 1:1 kafkaspout first bolt. i'm guessing #1 result of timeouts first bolt.
what want do: have kafkaspouts (limited 1 / kafka partition) able send bits random first bolt can run many more of these # of spouts. first , second bolts 1:1 spout first bolt should 1:many.
currently i'm using localorshufflegrouping connect between spout->bolt->bolt.
edit:
(re)reading storms docs see passage:
shuffle grouping: tuples randomly distributed across bolt's tasks in way such each bolt guaranteed equal number of tuples.
yet when @ load on executors first bolt see concentrated on 6 of them - seemingly ignoring other 24.
i'm missing large clue here.
Comments
Post a Comment