flink-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Juan Miguel Cejuela <jua...@tagtog.net>
Subject readFile, DataStream
Date Fri, 10 Nov 2017 11:54:23 GMT
Hi there,

I’m trying to watch a directory for new incoming files (with
StreamExecutionEnvironment#readFile) with a subsecond latency (interval
watch of ~100ms, and using the flag FileProcessingMode.PROCESS_CONTINUOUSLY
).

If many files come in within (under) the interval watching time, flink
doesn’t seem to get notice of the files, and as a result, the files do not
get processed. The behavior also seems undeterministic, as it likely
depends on timeouts and so on. For example, 10 new files come in
immediately (that is, essentially in parallel) and perhaps 3 files get
processed, but the rest 7 don’t.

I’ve extended and created my own FileInputFormat, for which I don’t do much
more than in the open function, log when a new file comes in. That’s how I
know that many fails get lost.

On the other hand, when I restart flink, *all* the files in the directory
are immediately processed. This is the expected behavior and works fine.

The situation of unprocessed files is a bummer.

Am I doing something wrong? Do I need to set something in the
configuration? Is it a bug in Flink?

Hopefully I described my problem clearly.

Thank you.
​

Mime
View raw message