Kafka running on Ceph
Hi All, Does anyone have any performance metrics running Kafka on Ceph? I briefly gathered at the 2016 Kafka Summit that there's an ongoing work between the Kafka community and RedHat in getting Kafka running successfully on Ceph. Is this correct? If so, what's timeline for that? Thanks Connie
Re: Support customized security protocol
@Ismael, what's the status of the SASL/PLAIN PR, https://github.com/apache/kafka/pull/341? On Tue, Jan 19, 2016 at 6:25 PM, tao xiaowrote: > The PR provides a new SASL mech but it doesn't provide a pluggable way to > implement user's own logic to do authentication. So I don't think the PR > will meet my need. > > I will write a KIP to open the discussion. > > p.s. Ismael, can you grant me the permission to create a KIP in Kafka > space? > > > On Wed, 20 Jan 2016 at 10:08 Ismael Juma wrote: > > > Hi Tao, > > > > The other way would be to implement a SASL provider: > > > > > > > https://docs.oracle.com/javase/8/docs/technotes/guides/security/sasl/sasl-refguide.html#PROV > > > > This would still require Kafka to be changed, some of the changes are in > > the following PR: > > > > https://github.com/apache/kafka/pull/341 > > > > As per the discussion in the PR above, a KIP is also required. > > > > Ismael > > > > On Wed, Jan 20, 2016 at 1:48 AM, tao xiao wrote: > > > > > Hi Ismael, > > > > > > BTW looks like I don't have the permission to add a KIP in Kafka space. > > Can > > > you please grant me the permission? > > > > > > On Wed, 20 Jan 2016 at 09:40 tao xiao wrote: > > > > > > > Hi Ismael, > > > > > > > > Thank you for your reply. I am happy to have a writeup on this. > > > > > > > > Can you think of any other ways to make security protocol pluggable > > > > instead of extending ChannelBuilder? > > > > > > > > On Wed, 20 Jan 2016 at 02:14 Ismael Juma wrote: > > > > > > > >> Hi Tao, > > > >> > > > >> As you say, security protocols are not currently pluggable. > > > >> `ChannelBuilder` is already an interface, but `SecurityProtocol` is > an > > > >> enum, which makes it hard for users to add additional security > > > protocols. > > > >> Changing this would probably require a KIP: > > > >> > > > >> > > > >> > > > > > > https://cwiki.apache.org/confluence/display/KAFKA/Kafka+Improvement+Proposals > > > >> > > > >> Ismael > > > >> > > > >> On Mon, Jan 18, 2016 at 3:15 AM, tao xiao > > wrote: > > > >> > > > >> > Hi Kafka team, > > > >> > > > > >> > I want to know if I can plug-in my own security protocol to Kafka > to > > > >> > implement project specific authentication mechanism. The current > > > >> supported > > > >> > authentication protocols, SASL/GSSAPI and SSL, are not supported > in > > my > > > >> > company and we have own security protocol to do authentication. > > > >> > > > > >> > Is it a good idea to make ChannelBuilder extensible so that I can > > > >> implement > > > >> > it with my own security channel? > > > >> > > > > >> > > > > > > > > > >
Re: [VOTE] 0.8.2.1 Candidate 1
+1 On Feb 18, 2015 7:23 PM, Matt Narrell matt.narr...@gmail.com wrote: +1 On Feb 18, 2015, at 7:56 PM, Jun Rao j...@confluent.io wrote: This is the first candidate for release of Apache Kafka 0.8.2.1. This only fixes one critical issue (KAFKA-1952) in 0.8.2.0. Release Notes for the 0.8.2.1 release https://people.apache.org/~junrao/kafka-0.8.2.1-candidate1/RELEASE_NOTES.html *** Please download, test and vote by Saturday, Feb 21, 7pm PT Kafka's KEYS file containing PGP keys we use to sign the release: http://kafka.apache.org/KEYS in addition to the md5, sha1 and sha2 (SHA256) checksum. * Release artifacts to be voted upon (source and binary): https://people.apache.org/~junrao/kafka-0.8.2.1-candidate1/ * Maven artifacts to be voted upon prior to release: https://repository.apache.org/content/groups/staging/ * scala-doc https://people.apache.org/~junrao/kafka-0.8.2.1-candidate1/scaladoc/ * java-doc https://people.apache.org/~junrao/kafka-0.8.2.1-candidate1/javadoc/ * The tag to be voted upon (off the 0.8.2 branch) is the 0.8.2.1 tag https://git-wip-us.apache.org/repos/asf?p=kafka.git;a=tag;h=c1b4c58531343dce80232e0122d085fc687633f6 /*** Thanks, Jun
Immediate Kafka security solution before 0.9 release
Hi All, Before Kafka 0.9 release is available, is there an immediate security solution that we can leverage? I've come across https://github.com/relango/kafka/tree/kafka_security and the IP address filter patch from Kafka 0.8.3, which has not have a set release date. Thanks, Connie
Re: How to recover from a disk full situation in Kafka cluster?
It looks like org.apache.kafka.clients.producer.KafkaProducer is not available in 0.8.1.1 client jar. So, we'll stay with kafka.javaapi.producer.Producer implementation. Thanks, Connie On Fri, Jul 18, 2014 at 5:13 PM, Neha Narkhede neha.narkh...@gmail.com wrote: One option is to reduce the value of topic.metadata.refresh.interval.ms but the concern is that may end up sending too many requests to the brokers causing overhead. I'd suggest you use the new producer under org.apache.kafka.clients.producer.KafkaProducer that does not have the problem. It is fairly new but has gone through some level of testing now and we will appreciate any feedback/bugs that you can report back. Thanks, Neha On Fri, Jul 18, 2014 at 4:23 PM, Connie Yang cybercon...@gmail.com wrote: Sure, I will try to take a snapshot of the data distribution when it happens next time. Assuming the topic.metadata.refresh.interval.ms is the concern, how should we unstuck our producers? The important note from that the documentation seems to suggest that the metadata refresh will only happen AFTER the message is sent. The producer generally refreshes the topic metadata from brokers when there is a failure (partition missing, leader not available...). It will also poll regularly (default: every 10min so 60ms). If you set this to a negative value, metadata will only get refreshed on failure. If you set this to zero, the metadata will get refreshed after each message sent (not recommended). Important note: the refresh happen only AFTER the message is sent, so if the producer never sends a message the metadata is never refreshed Thanks, Connie On Fri, Jul 18, 2014 at 3:58 PM, Neha Narkhede neha.narkh...@gmail.com wrote: Does this mean that we should set auto.leader.rebalance.enable to true? I wouldn't recommend that just yet since it is not known to be very stable. You mentioned that only 2 brokers ever took the traffic and the replication factor is 2, makes me think that the producer stuck to 1 or few partitions instead of distributing the data over all the partitions. This is a known problem in the old producer where the default value of a config ( topic.metadata.refresh.interval.ms), that controls how long a producer sticks to certain partitions, is 10 mins. So it effectively does not distribute data evenly across all partitions. If you see the same behavior next time, try to take a snapshot of data distribution across all partitions to verify this theory. Thanks, Neha On Thu, Jul 17, 2014 at 5:43 PM, Connie Yang cybercon...@gmail.com wrote: It might appear that the data is not balanced, but it could be as a result of the imbalanced leaders setting. Does this mean that we should set auto.leader.rebalance.enable to true? Any other configuration we need to change as well? As I mentioned before, we use pretty much use the default setting. All of our topics have replication factor of 2 (aka 2 copies per message). We don't have the topic output when we had the problem, but here's our topic output after we ran the kafka-preferred-replica-election.sh tool as suggested: $KAFKA_HOME/bin/kafka-topics.sh --zookeeper zkHost1:2181,zkHost2:2181,zkHost3:2181 --describe --topic=myKafkaTopic Topic:myKafkaTopic PartitionCount:24 ReplicationFactor:2 Configs: retention.ms=4320 Topic: myKafkTopic Partition: 0 Leader: 2 Replicas: 2,1 Isr: 1,2 Topic: myKafkTopic Partition: 1 Leader: 3 Replicas: 3,2 Isr: 3,2 Topic: myKafkTopic Partition: 2 Leader: 4 Replicas: 4,3 Isr: 3,4 Topic: myKafkTopic Partition: 3 Leader: 5 Replicas: 5,4 Isr: 5,4 Topic: myKafkTopic Partition: 4 Leader: 6 Replicas: 6,5 Isr: 5,6 Topic: myKafkTopic Partition: 5 Leader: 7 Replicas: 7,6 Isr: 6,7 Topic: myKafkTopic Partition: 6 Leader: 8 Replicas: 8,7 Isr: 7,8 Topic: myKafkTopic Partition: 7 Leader: 9 Replicas: 9,8 Isr: 9,8 Topic: myKafkTopic Partition: 8 Leader: 10 Replicas: 10,9 Isr: 10,9 Topic: myKafkTopic Partition: 9 Leader: 11 Replicas: 11,10 Isr: 11,10 Topic: myKafkTopic Partition: 10 Leader: 12 Replicas: 12,11 Isr: 11,12 Topic: myKafkTopic Partition: 11 Leader: 13 Replicas: 13,12 Isr: 12,13 Topic: myKafkTopic Partition: 12 Leader: 14 Replicas: 14,13 Isr: 14,13 Topic: myKafkTopic Partition: 13 Leader: 15 Replicas: 15,14 Isr: 14,15 Topic: myKafkTopic Partition: 14 Leader: 16 Replicas: 16,15 Isr: 16,15 Topic: myKafkTopic Partition: 15 Leader: 17 Replicas: 17,16 Isr: 16,17 Topic: myKafkTopic Partition: 16 Leader: 18 Replicas: 18,17 Isr: 18,17 Topic: myKafkTopic Partition: 17 Leader: 19 Replicas: 19,18 Isr: 18,19 Topic: myKafkTopic Partition: 18 Leader: 20 Replicas: 20,19 Isr: 20,19 Topic: myKafkTopic Partition: 19 Leader: 21 Replicas: 21,20 Isr: 20,21 Topic
Re: How to recover from a disk full situation in Kafka cluster?
Sure, I will try to take a snapshot of the data distribution when it happens next time. Assuming the topic.metadata.refresh.interval.ms is the concern, how should we unstuck our producers? The important note from that the documentation seems to suggest that the metadata refresh will only happen AFTER the message is sent. The producer generally refreshes the topic metadata from brokers when there is a failure (partition missing, leader not available...). It will also poll regularly (default: every 10min so 60ms). If you set this to a negative value, metadata will only get refreshed on failure. If you set this to zero, the metadata will get refreshed after each message sent (not recommended). Important note: the refresh happen only AFTER the message is sent, so if the producer never sends a message the metadata is never refreshed Thanks, Connie On Fri, Jul 18, 2014 at 3:58 PM, Neha Narkhede neha.narkh...@gmail.com wrote: Does this mean that we should set auto.leader.rebalance.enable to true? I wouldn't recommend that just yet since it is not known to be very stable. You mentioned that only 2 brokers ever took the traffic and the replication factor is 2, makes me think that the producer stuck to 1 or few partitions instead of distributing the data over all the partitions. This is a known problem in the old producer where the default value of a config ( topic.metadata.refresh.interval.ms), that controls how long a producer sticks to certain partitions, is 10 mins. So it effectively does not distribute data evenly across all partitions. If you see the same behavior next time, try to take a snapshot of data distribution across all partitions to verify this theory. Thanks, Neha On Thu, Jul 17, 2014 at 5:43 PM, Connie Yang cybercon...@gmail.com wrote: It might appear that the data is not balanced, but it could be as a result of the imbalanced leaders setting. Does this mean that we should set auto.leader.rebalance.enable to true? Any other configuration we need to change as well? As I mentioned before, we use pretty much use the default setting. All of our topics have replication factor of 2 (aka 2 copies per message). We don't have the topic output when we had the problem, but here's our topic output after we ran the kafka-preferred-replica-election.sh tool as suggested: $KAFKA_HOME/bin/kafka-topics.sh --zookeeper zkHost1:2181,zkHost2:2181,zkHost3:2181 --describe --topic=myKafkaTopic Topic:myKafkaTopic PartitionCount:24 ReplicationFactor:2 Configs: retention.ms=4320 Topic: myKafkTopic Partition: 0 Leader: 2 Replicas: 2,1 Isr: 1,2 Topic: myKafkTopic Partition: 1 Leader: 3 Replicas: 3,2 Isr: 3,2 Topic: myKafkTopic Partition: 2 Leader: 4 Replicas: 4,3 Isr: 3,4 Topic: myKafkTopic Partition: 3 Leader: 5 Replicas: 5,4 Isr: 5,4 Topic: myKafkTopic Partition: 4 Leader: 6 Replicas: 6,5 Isr: 5,6 Topic: myKafkTopic Partition: 5 Leader: 7 Replicas: 7,6 Isr: 6,7 Topic: myKafkTopic Partition: 6 Leader: 8 Replicas: 8,7 Isr: 7,8 Topic: myKafkTopic Partition: 7 Leader: 9 Replicas: 9,8 Isr: 9,8 Topic: myKafkTopic Partition: 8 Leader: 10 Replicas: 10,9 Isr: 10,9 Topic: myKafkTopic Partition: 9 Leader: 11 Replicas: 11,10 Isr: 11,10 Topic: myKafkTopic Partition: 10 Leader: 12 Replicas: 12,11 Isr: 11,12 Topic: myKafkTopic Partition: 11 Leader: 13 Replicas: 13,12 Isr: 12,13 Topic: myKafkTopic Partition: 12 Leader: 14 Replicas: 14,13 Isr: 14,13 Topic: myKafkTopic Partition: 13 Leader: 15 Replicas: 15,14 Isr: 14,15 Topic: myKafkTopic Partition: 14 Leader: 16 Replicas: 16,15 Isr: 16,15 Topic: myKafkTopic Partition: 15 Leader: 17 Replicas: 17,16 Isr: 16,17 Topic: myKafkTopic Partition: 16 Leader: 18 Replicas: 18,17 Isr: 18,17 Topic: myKafkTopic Partition: 17 Leader: 19 Replicas: 19,18 Isr: 18,19 Topic: myKafkTopic Partition: 18 Leader: 20 Replicas: 20,19 Isr: 20,19 Topic: myKafkTopic Partition: 19 Leader: 21 Replicas: 21,20 Isr: 20,21 Topic: myKafkTopic Partition: 20 Leader: 22 Replicas: 22,21 Isr: 22,21 Topic: myKafkTopic Partition: 21 Leader: 23 Replicas: 23,22 Isr: 23,22 Topic: myKafkTopic Partition: 22 Leader: 24 Replicas: 24,23 Isr: 23,24 Topic: myKafkTopic Partition: 23 Leader: 1 Replicas: 1,24 Isr: 1,24 Thanks, Connie On Thu, Jul 17, 2014 at 4:20 PM, Neha Narkhede neha.narkh...@gmail.com wrote: Connie, After we freed up the cluster disk space and adjusted the broker data retention policy, we noticed that the cluster partition was not balanced based on topic describe script came from Kafka 0.8.1.1 distribution. When you say the cluster was not balanced, did you mean the leaders or the data? The describe topic tool does not give information about data sizes, so I'm assuming you are referring to leader imbalance. If so, the right tool to run is kafka-preferred-replica-election.sh not partition reassignment. In general, assuming the partitions were
What happens to Kafka when ZK lost its quorum?
Hi all, Can Kafka producers, brokers and consumers still be processing messages and functioning in their normal states if Zookeeper lost its quorum? Thanks, Connie
What happens to Kafka when ZK lost its quorum or becomes unstable?
Hi, Can the producers, brokers and consumers still be processing messages when their ZK cluster lost its quorum or becomes unstable? I know this is rather general question as it may depends on what configuration these use. So, please enumerate all of those combinations. Thanks, Connie