Table of Contents |
---|
Status
Current state: Under Discussion
Discussion thread: here
JIRA:
Motivation
People use Kafka, among many reasons, because they need to be sure their messages are correctly processed by their applications. Classic configuration is to have 3 replica, and commit the offset of a message once it has been correctly processed. Developers use this configuration because it is important not to lose any messages. More important that not losing messages is to be notified when a message is lost without being consumed.
There are some situations where messages are lost silently:
...
Messages stored in Kafka will expire and disappear silently based on retention time and size configuration. Consumers have no way to know they have missed messages.
I propose to log expose a warning JMX metric after a message has been removed due to topic time/size retention settings, for a set of consumer groups specified on the topic configuration.
The This could be implemented because the kafka brokers know the information needed to achieve for the goaltask:
- offset of the message that has been removed.
- last offset consumed from a consumer group.
The compacted topics are out of scope, because when users chose compacted topics they are interested on the last value of they key, not on the intermediary states.
Public Interfaces
A new property at topic level property would should be created:
...
Name | Description | Type | Default | Valid Values | Server default property | Importance |
---|---|---|---|---|---|---|
non.consumed.offsets.groups | comma separated list of consumer groups that will |
...
Proposed Changes
The modifications introduced are in blue:
- The LogManager variable logsToBeDeleted contains logs and time schedule for deletion.
...
expose a metric with the number of messages that expired before being consumed | List | "" | "" | medium |
A new JMX metric should be created to be exposed by the broker:
Metric / Attribute Name | Description | MBEAN NAME |
---|---|---|
non-consumed-total | Number of messages expired without being consumed by a consumer group, per topic and partition |
Proposed Changes
Currently the LogManager schedule the "kafka-delete-logs" thread, that will call the deleteLogs() method. Is possible to add into that method the metric to expose the number of offsets non consumed by a list of consumer groups.
The pseudo code starts on the line 19:
Code Block | ||||||
---|---|---|---|---|---|---|
| ||||||
private def deleteLogs(): Unit = {
var nextDelayMs = 0L
try {
def nextDeleteDelayMs: Long = {
if (!logsToBeDeleted.isEmpty) {
val (_, scheduleTimeMs) = logsToBeDeleted.peek()
scheduleTimeMs + currentDefaultConfig.fileDeleteDelayMs - time.milliseconds()
} else
currentDefaultConfig.fileDeleteDelayMs
}
while ({nextDelayMs = nextDeleteDelayMs; nextDelayMs <= 0}) {
val (removedLog, _) = logsToBeDeleted.take()
if (removedLog != null) {
try {
removedLog.delete()
info(s"Deleted log for partition ${removedLog.topicPartition} in ${removedLog.dir.getAbsolutePath}.")
//
// KIP-490: log when consumer groups lose a message because offset has been deleted
//
val consumerGroupsSubscribed : Seq[String] = getConsumerGroups(removedLog.topicPartition.topic() );
val groupsToNotify : Seq[String] = consumerGroupsSubscribed intersect groupsTobeNotify // value get from topic config property 'retention.notify.groups'
groupsToNotify.forEach( {
val lastCosumedOffsetGroup : Integer = getLastOffsetConsumed( _, removedLog.topicPartition);
if(lastCosumedOffsetGroup < removedLog.nextOffsetMetadata) {
// increment and expose JMS metric non-consumed-total.
}
})
} catch {
case e: KafkaStorageException =>
error(s"Exception while deleting $removedLog in dir ${removedLog.dir.getParent}.", e)
}
}
}
} catch {
case e: Throwable =>
error(s"Exception in kafka-delete-logs thread.", e)
} finally {
try {
scheduler.schedule("kafka-delete-logs",
deleteLogs _,
delay = nextDelayMs,
unit = TimeUnit.MILLISECONDS)
} catch {
case e: Throwable =>
if (scheduler.isStarted) {
// No errors should occur unless scheduler has been shutdown
error(s"Failed to schedule next delete in kafka-delete-logs thread", e)
}
}
}
} |
...
Compatibility, Deprecation, and Migration Plan
There is no impact on existing features.
Rejected Alternatives
Logging on broker side
To write logs on the broker to side to alert that messages have not been consumed has been rejected in favour of standard monitoring. Main reason is the big amount of data that could be generated.