We are running Kafka 0.7.2. We set log.roll.hours=1. I hoped that meant logs would be rolled every hour, or more. Only, sometimes logs that are many hours (sometimes days) old have more data added to them. This perturbs our systems for reasons I won't get in to.
Have others observed this? Is it a bug? Is there a planned fix?
I don't know Scala or Kafka well, but I have proposal for why this might happen: upon restart, a broker forgets when its log files have been appended to ("firstAppendTime"). Then a potentially infinite amount of time later, the restarted broker receives another message for the particular (topic, partition), and starts the clock again. It will then roll over that log after an hour.
That logic in 0.7.2 seems correct. Basically, firstAppendTime is set on first append to a log segment. Then, later on, when a new message is appended and the elapsed time since firstAppendTime is larger than the roll time, a new segment is rolled. Is your data constantly being produced?
Jun On Thu, Apr 25, 2013 at 12:44 PM, Dan Frankowski <[EMAIL PROTECTED]> wrote:
We have high-volume topics and low-volume topics. The problem occurs more often for low-volume topics to be sure.
But if my hypothesis is correct about why it is happening, here is a case where rolling is longer than an hour, even on a high volume topic:
- write to a topic for 20 minutes - restart the broker - wait for 5 days - write to a topic for 20 minutes - restart the broker - write to a topic for an hour
The rollover time was now 5 days, 1 hour, 40 minutes. You can make it as long as you want. Does this make sense?
We would like the rollover time to be no more than an hour, even if the broker is restarted, or the topic is low-volume.
The cleanest way to do that might be to roll over on the hour no matter when the file was started. That would be too fast sometimes, but that's fine. A second way would be to embed the first append time in the file name. A third way (not perfect, but an approximation at least) would be to not to write to a segment if firstAppendTime is not defined and the timestamp on the file is more than an hour old. There are probably other ways.
What say you? On Thu, Apr 25, 2013 at 9:49 PM, Jun Rao <[EMAIL PROTECTED]> wrote:
Yes, for low volume topic, the time-based rolling can be imprecise. Could you file a jira and describe your suggestions there? Ideally, we should set firstAppendTime to the file creation time. However, it doesn't seem you can get the creation time in java.
Jun On Thu, Apr 25, 2013 at 11:12 PM, Dan Frankowski <[EMAIL PROTECTED]> wrote:
It's possible to find this setting (and several other undocumented settings) by looking at the source code. I'm just not sure why the complete set of options is not documented on the site (is it meant to be experimental?).
Jason On Fri, Apr 26, 2013 at 8:19 AM, Dan Frankowski <[EMAIL PROTECTED]> wrote:
I don't know how Kafka's rollover algorithm is implemented, but this is common behavior for other logging frameworks. You would need a separate watcher/scheduled thread to rollover a log file, even if no events were coming in. Logback (and probably log4j, by the same author) dispenses with the watcher thread. Instead, it checks each message as it comes in and decides whether the message should belong in a new file. If it should, a rollover of the old file is triggered and the message is deposited in the new file. But no rollover will occur until a message that belongs in a new file arrives.
On Fri, Apr 26, 2013 at 9:52 AM, Jason Rosenberg <[EMAIL PROTECTED]> wrote:
I believe there is a separate watcher thread. The only issue is upon restart the broker forgets when the file was created. The behavior I described (files can be appended to infinitely) is awkward for us. We have tried to work around it. On Fri, Apr 26, 2013 at 10:32 AM, Adam Talaat <[EMAIL PROTECTED]> wrote:
@Dan: Upon restart of the broker, if a segment already has data, the broker resets the firstAppendTime of the segment to the time when that segment's file handles are being loaded into memory. Thus as you correctly explained, every time you shut down a broker, the broker essentially forgets the firstAppendTime. This behavior is present in both 0.7.2 and 0.8.
As Jun said, ideally we should set firstAppendTime to the file creation time. Unfortunately Java nio can provide you provide you that information only if the underlying filesystem implementation supports the notion of file creation time.
Thanks for filing the JIRA, these are good suggestions.
@Jason: Thanks for pointing out that log.roll.hours is not documented on the website. This config was added late in 0.7 and we probably forgot to update the website. We have filed KAFKA-834/KAFKA-835 to update the configs and other documentation on the website in general. Please let us know if you see any other missing piece.
On 4/27/13 2:36 PM, "Dan Frankowski" <[EMAIL PROTECTED]> wrote:
Someone pointed out a particularly easy fix: don't reuse files after a restart. Done. I really like that. Simple. Any chance of this happening any time soon? On Sun, Apr 28, 2013 at 2:04 AM, Swapnil Ghike <[EMAIL PROTECTED]> wrote:
Jun On Thu, May 2, 2013 at 2:22 PM, Dan Frankowski <[EMAIL PROTECTED]> wrote:
NEW: Monitor These Apps!
Apache Lucene, Apache Solr and all other Apache Software Foundation projects and their respective logos are trademarks of the Apache Software Foundation.
Elasticsearch, Kibana, Logstash, and Beats are trademarks of Elasticsearch BV, registered in the U.S. and in other countries. This site and Sematext Group is in no way affiliated with Elasticsearch BV.
Service operated by Sematext