Current state: Under Discussion

Discussion thread: 

JIRA: KAFKA-14127 - Getting issue details... STATUS

Please keep the discussion on the mailing list rather than commenting on the wiki (wiki discussions get unwieldy fast).


Support for multiple log directories per broker, aka JBOD (Just a Bunch Of Disks) came in KIP-112 and since then JBOD has been an important feature in Kafka, allowing it to run on large deployments with multiple storage devices per broker.

To ensure availability, when a partition leader fails, the controller should elect a new leader from one of the other in-sync replicas. But the controller does not check whether each leader is correctly performing its duties, instead the controller simply assumes that each broker is working correctly if it is still an active member of the cluster. In KRaft, cluster membership is based on timely heartbeat requests sent by each broker to the active controller. In ZooKeeper, cluster membership is based on an ephemeral zNode under /brokers/ids.

In KRaft mode, when a single log directory fails, the broker will be unable to be either a leader or a follower for any partitions in that log directory, but the controller will have no signal that it needs to update leadership and ISR for the replicas in that log directory, as the broker will continue to send heartbeat requests. 1

In ZooKeeper mode when a log directory fails, the broker sends a notification to the controller which then sends a full LeaderAndIsr request to the broker, listing all the partitions for all log directories for that broker. The controller relies on per-partition error results from the broker to update leadership and ISR for the replicas in the failed log directory. Without this notification, the partitions with leadership on that log directory will not get a new leader assigned and would remain unavailable.

Support for KRaft in JBOD, was proposed and accepted back in KIP-589 — with a new RPC from the broker to the controller indicating the affected topic partitions in a failed log directory — but the implementation was never merged and concerns were raised with possible large requests from the broker to the controller.

KIP-833 was accepted, with plans to mark KRaft as production ready and deprecate ZooKeeper mode, but JBOD is still a missing feature in KRaft. This KIP aims to provide support for JBOD in KRaft, while avoiding any RPC having to list all the partitions in a log directory.

Public interfaces

Command line tools

The format  sub-command in the  tool will produce  in a new format described in the following section which includes two new properties and directory.ids. This sub-command already supports a formatting more than one log directory — by expecting a list of configured log.dirs  —  and "formatting" only the ones that need so. All configured log directories must be available for format to run.

The  version field will be bumped from 1 to 2. Two new properties and directory.ids will be added to the file in each log directory, including the metadata.log.dir . The first property, indicates the UUID for the log directory where the file is located, the second property, directory.ids  lists all the UUIDs for all the configured log directories. If the  file doesn't exist for the metadata.log.dir  the Kafka node will fail to start. If the  file exists but it doesn't contain these two properties a new one will be generated and the  files will be updated. The  tool will be extended to generate or update the two properties as described in the previous section.

Metadata records

RegisterBrokerRecord and BrokerRegistrationChangeRecord will both have two new fields:

{ "name": "OnlineLogDirs", "type":  "[]uuid", "versions":  "2+",
"about": "Log directories configured in this broker which are available." },
{ "name": "OfflineLogDirs", "type": "[]uuid", "versions": "2+",
"about": "Log directories configured in this broker which are not available." }

PartitionRecord and PartitionChangeRecord will both have a new Assignment field which will replace the current Repicas field:

{ "name": "Replicas", "type":  "[]int32", "versions":  "0", "entityType": "brokerId",
"about": "The replicas of this partition, sorted by preferred order." },
{ "name": "Assignment", "type": "[]ReplicaAssignment", "versions": "1+",
"about": "The replicas of this partition, sorted by preferred order.", "fields": [
{ "name": "Broker", "type": "int32", "versions": "1+", "entityType": "brokerId",
"about": "The broker ID hosting the replica." },
{ "name": "Directory", "type": "uuid", "versions": "1+",
"about": "The log directory hosting the replica" }

RPC requests

BrokerRegistrationRequest will include the following two new fields:

{ "name": "OnlineLogDirs", "type":  "[]uuid", "versions":  "1+",
"about": "Log directories configured in this broker which are available." },
{ "name": "OfflineLogDirs", "type": "[]uuid", "versions": "1+",
"about": "Log directories configured in this broker which are not available." }

BrokerHeartbeatRequest will include the following new field:

{ "name": "LogDirsOfflined", "type":  "[]uuid", "versions":  "1+",
"about": "Log directories that failed and went offline." }

A new RPC named AssignReplicasToDirs will be introduced with the following request and response:

"apiKey": <TBD>,
"type": "request",
"listeners": ["controller],
"name": "AssignReplicasToDirsRequest",
"validVersions": "0",
"flexibleVersions": "0+",
"fields": [
{ "name": "BrokerId", "type": "int32", "versions": "0+", "entityType": "brokerId",
"about": "The ID of the requesting broker" },
{ "name": "BrokerEpoch", "type": "int64", "versions": "0+", "default": "-1",
"about": "The epoch of the requesting broker" },
{ "name": "Directories", "type": "[]DirectoryData", "versions": "0+", "fields": [
{ "name": "Id", "type": "uuid", "versions": "0+", "about": "The ID of the directory" },
{ "name": "Topics", "type": "[]TopicData", "versions": "0+", "fields": [
{ "name": "Name", "type": "string", "versions": "0+", "entityType": "topicName",
"about": "The name of the assigned topic" },
{ "name": "Partitions", "type": "[]PartitionData", "versions": "0+", "fields": [
{ "name": "PartitionIndex", "type": "int32", "versions": "0+",
"about": "The partition index" }
"apiKey": <TBD>,
"type": "response",
"name": "AssignReplicasToDirsResponse",
"validVersions": "0",
"flexibleVersions": "0+",
"fields": [
{ "name": "ThrottleTimeMs", "type": "int32", "versions": "0+",
"about": "The duration in milliseconds for which the request was throttled due to a quota violation, or zero if the request did not violate any quota." },
{ "name": "ErrorCode", "type": "int16", "versions": "0+",
"about": "The top level response error code" },
{ "name": "Directories", "type": "[]DirectoryData", "versions": "0+", "fields": [
{ "name": "Id", "type": "uuid", "versions": "0+", "about": "The ID of the directory" },
{ "name": "Topics", "type": "[]TopicData", "versions": "0+", "fields": [
{ "name": "Name", "type": "string", "versions": "0+", "entityType": "topicName",
"about": "The name of the assigned topic" },
{ "name": "Partitions", "type": "[]PartitionData", "versions": "0+", "fields": [
{ "name": "PartitionIndex", "type": "int32", "versions": "0+",
"about": "The partition index" },
{ "name": "ErrorCode", "type": "int16", "versions": "0+",
"about": "The partition level error code" }

Proposed changes


MBean nameDescription
kafka.server:type=KafkaServer,name=NumMismatchingReplicaToLogDirAssignmentsThe number of replicas hosted by the broker that are either missing a log directory assignment in the cluster metadata or are currently found in a different log directory. 

Storage format command

The format subcommand will be updated to ensure each log directory has an assigned UUID and it will persist two new properties in the  file:

  • A property named indicating the UUID for the log directory where the file is located. The value is base64 encoded, like the cluster UUID.
  • A property named directory.ids indicating the complete list of all UUIDs for each configured log directory. Values are base64 encoded and comma-separated. The order does not matter.

The  version field will be bumped from 1 to 2.

The UUIDs for each log directory are automatically generated by the tool if there isn't one assigned already in an existing  file.

Having a persisted UUID at the root of each log directory allows the broker to identify the log directory regardless of the mount path.
Having a persisted list of all UUIDs for all configured log directories allows the broker to determine the UUIDs of unavailable (offline) log directories, as the  files for the offline log directories are likely to be unavailable.


Given the following

(... other non interesting properties omitted ...)

The command ./bin/ format -c /tmp/ --cluster-id 41QSStLtR3qOekbX4ZlbHA  would generate three  files that could look like the following:

/var/lib/kafka/metadata/ :

#Thu Aug 18 15:23:07 BST 2022
/mnt/d1/ :
#Thu Aug 18 15:23:07 BST 2022
/mnt/d2/ :
#Thu Aug 18 15:23:07 BST 2022

Each directory, including the directory that holds the cluster metadata topic — metadata.log.dir  — has a different and respective value as the directory ID. The full set of directory IDs — for all log dirs in log.dirs  but also metadata.log.dir — is persisted in all three metadata files.

In the example above, we can identify the following directory mapping:

  • /var/lib/kafka/metadata  has log directory UUID e6umYSUsQyq7jUUzL9iXMQ 
  • /mnt/d1  has log directory UUID b4d9ExdORgaQq38CyHwWTA 
  • /mnt/d2 has log directory UUID P2aL9r4sSqqyt7bC0uierg 

If some but not all log directories are unavailable, the broker is able to identify which UUIDs refer to offline log directories. 


Broker lifecycle management

When the broker starts up and initializes LogManager, for each configured log directory (in log.dirs ) it will load the UUID for each log directory ( ) and the list of all log directory UUIDs (directory.ids), by reading the file at the root of each log directory.

  • If there are any two log directories with the same UUID, the broker will fail at startup
  • If there are any files missing, a new UUID is generated, and assigned to that log directory by updating the file
  • If there are no offline log directories the broker will also create or amend the directory.ids field in each file as required

If there are offline log directories, the broker might not be able to determine the UUID for each specific offline log directory, but by diffing diffing directory.ids with the loaded UUIDs from all the set of offline log directory UUIDs can still be determined.

After loading the broker will diff all the UUIDs in  with the full set of all UUIDs (in directory.ids) to obtain the set of UUIDs for offline log directories. The sets of both online and offline log directory UUIDs are sent along in the broker registration request to the controller. If log directory that holds the cluster metadata topic is configured separately to a different path — using metadata.log.dir — then the respective UUID for this log directory is excluded from both online and offline sets, as the broker cannot run if this particular log directory is unavailable.

If a new entry is added in the log.dirs  configuration, the broker can always expand directory.ids as it can determine the "set of UUIDs for online log directories" + "set of UUIDs for offline log directories" + newly generated UUID for the log directory.

If an entry is removed from log.dirs  the broker can also automatically update directory.ids as long as no log directories are offline when the broker comes back up. The broker will need to be able to access all to determine the new full set of UUIDs. An unresolvable mismatch might occur if some log directory was removed from log.dirs , and some other log directory is offline. It is not possible to determine which UUID belonged to each of the missing log dirs. The UUID for the removed log directory needs to be removed from directory.ids  but the UUID for the offline log directory should stay. Upon an unresolvable mismatch between the number of entries configured in log.dirs  and found in  under directory.ids the broker will fail at startup.

Replica management

As the broker catches up with metadata, and sees the partitions which it should be hosting, it will check the associated log directory UUID for each partition.

  • If the partition is not assigned to a log directory (refers to Uuid.ZERO)
    • If the partition already exists, the broker uses the new RPC — AssignReplicasToDirs — to notify the controller to change the metadata assignment to the actual log directory.
    • If the partition does not exist, the broker selects a log directory and uses the new RPC — AssignReplicasToDirs — to notify the controller to create the metadata assignment to the actual log directory.
  • If the partition is assigned to an online log directory
    • If the partition does not exist it is created in the indicated log directory.
    • If the partition already exists in the indicated log directory and no future replica exists, then no action is taken.
    • If the partition already exists in the indicated log directory, and there is a future replica in another log directory, then the broker starts the process to replicate the current replica to the future replica.
    • If the partition already exists in another online log directory and is a future replica in the log directory indicated by the metadata, the broker will replace the current replica with the future replica after making sure that the future replica as fully caught up with the current replica.
    • If the partition already exists in another online log directory, the broker uses the new RPC — AssignReplicasToDirs — to the controller to change the metadata assignment to the actual log directory. The partition might have been moved to a different log directory. whilst the broker was offline. 
    • If the broker knows that the partition already exists in a different log directory that is now offline, then the controller might not have known to update leadership and ISR when the log directory failure was communicated to the controller – unlikely, but possibly due to some synchronisation failure in the replica to log directory assignment between the broker and the metadata. To prevent the partition from remaining offline, the broker uses AssignReplicasToDirs to change the metadata assignment to an offline log directory. The controller learns the replica is offline and updates leadership and ISR.
  • If the partition is assigned to an offline log directory, no action is taken — the controller is already aware of this, and we don't want to fill the remaining online log directories with replicas that existed in the offline ones.
  • If the partition is assigned to an unknown log directory, no action is taken — the controller is already aware of this and will reassign the replica to one of the online log directories in a future metadata update. 

If the broker is configured with multiple log directories it remains FENCED until it can verify that all partitions are assigned to the correct log directories in the cluster metadata. This excludes the log directory that hosts the cluster metadata topic, if it is configured separately to a different path — using metadata.log.dir.

When replicas are moved between directories, using the existing AlterReplicaLogDirs RPC, the receiving broker will start moving the replicas using AlterReplicaLogDirs threads as usual. When a future replica first catches out the broker will asynchronously communicate the log directory change to the controller using the new RPC – AssignReplicasToDirs  – but keep the AlterReplicaLogDirs thread going. Once the broker receives confirmation of the metadata change then it briefly blocks appends to the old replica, makes sure the future log fully caches up and makes the switch. By delaying the metadata change until the future replica has caught up we minimize the chance of a log directory failure happening with an incorrect replica to log directory assignment in the metadata.

Metadata caching

Replicas are considered offline if the replica references a log directory which is not in the list of online log directories for the broker ID hosting the replica.

Handling log directory failures

When one or more log directories become offline, the broker will communicate this change using the new field LogDirsOfflined  in the BrokerHeartbeat  request — indicating the UUIDs of the new offline log directories. The UUIDs for the newly failed log directories are included in the BrokerHeartbeat  request until the broker receives a successful response.

If a log directory fails and the active controller learns of the failed log directory at a time when the replica-to-log-directory assignment in the metadata is incorrect then the controller might not know to update the leadership and ISR for some replicas, if some of those replicas are leaders then those partitions will become unavailable. As a fallback safety mechanism in case of any synchronisation issues, when a log directory fails, the broker will check log directory assignments in the metadata and if there are any missing replicas, the broker will use AlterReplicaLogDirs assigning the replicas to the offline log directory, to convey to the controller that these replicas are offline and need leadership and ISR updates. 


Replica placement

For any new partitions, the active controller will use Uuid.ZERO as the initial value for log directory UUID for each replica. Each broker hosting replicas then assigns a log directory UUID and communicates it back to the active controller using the new RPC AssignReplicasToDirs so that cluster metadata can be updated with the log directory assignment.

Handling log directory failures

When a controller receives a BrokerHeartbeat request from a broker that indicates any UUIDs under the new LogDirsOfflined field, it will:

  • Persist a BrokerRegistrationChange record, with the new list of online and offline log directories
  • Update the Leader and ISR for all the replicas assigned to the failed log directories, persisting PartitionChangeRecords, in similar way to how leadership and ISR is updated when a broker becomes fenced, unregistered or shuts down.

Handling replica assignments

The controller accepts the AssignReplicasToDirs RPC and persists the assignment into metadata records. If the indicated log directory UUID is not a registered log directory then the call fails with error 57 — LOG_DIR_NOT_FOUND . If the indicated log directory UUID is listed as offline, then the replica is considered offline and the leader and ISR is updated accordingly, same as when the BrokerHeartbeat indicates a new offline log directory.

Broker registration

Upon a broker registration request the controller will persist the broker registration as cluster metadata including the online and offline log directories for that broker. The controller may receive a new list of online and offline log directories — different from what was previously persisted in the cluster metadata for the requesting broker.

  • If there are no indicated online log directory UUIDs the request is invalid and the controller replies with an error — INVALID_REQUEST.
  • If there are any missing log directories this means those have been removed from the broker’s configuration, so the controller will reassign all replicas currently assigned to the missing log directories to Uuid.ZERO to delegate the choice of log directory the broker, which will then report the choice via the AssignReplicasToDirs RPC.
  • If multiple log directories are registered the broker will remain fenced until the controller learns of all the partition to log directory placements in that broker - i.e. no remaining replicas assigned to Uuid.ZERO . The broker will indicate these using the AssignReplicasToDirs RPC.

    • The broker remains fenced by not wanting to unfence itself in heartbeat requests until the number of mismatching replica to log directory assignments is zero. This number is represented by the new metric NumMismatchingReplicaToLogDirAssignments.
  • If multiple log directories are registered and some of them are new (not present in previous registration) then these log directories are assumed to be empty. If they are not, the broker will use the AssignReplicasToDirs  RPC to correct assignment and choose not to become UNFENCED before the metadata is correct.

Brokers whose registration indicates that multiple log directories are configured remain FENCED until all log directory assignments for that broker are learnt by the active controller and persisted into metadata.

Compatibility, Deprecation, and Migration Plan

The metadata.version will be bumped to gate changes to the RPCs and metadata records.

Migrating a cluster in KRaft without JBOD

The cluster needs to be upgraded before configuring multiple entries in log.dirs. As the upgraded brokers come up, the existing  files in each broker are updated with a generated  and directory.ids . After the upgrade, the metadata.version feature flag needs to be upgraded using Then the brokers can be reconfigured with multiple entries in log.dirs.

Upon being reconfigured with multiple log directories, brokers will update and generate as necessary to reflect the new log directories. Brokers will then register the log directories with the controller via BrokerRegistration and use AssignReplicasToDirs to create the partition-logdirectory assignments in the cluster metadata before becoming UNFENCED.

Migrating a cluster in ZK mode running with JBOD

Migration into KRaft mode is addressed in KIP-866. That migration is extended in the following way:

  • As per KIP-866, a separate Controller quorum is setup first, and only then the existing brokers are reconfigured and upgraded.
  • When configured for the migration and while still in ZK mode, brokers will:
    • update to generate and include  and directory.ids;
    • send BrokerRegistrationRequest including the log directory UUIDs;
    • notify the controller of log directory failures via BrokerHeartbeatRequest.
  • During the migration, the controller:
    • persists log directories indicated in broker registration requests in the cluster metadata;
    • relies on heartbeat requests to detect log directory failure instead of monitoring the ZK znode for notifications;
    • still uses full LeaderAndIsr requests to process log directory failures for any brokers still running in ZK mode.
  • The brokers restarting into KRaft mode will want to stay fenced until their log directory assignments for all hosted partitions are persisted in the cluster metadata.
  • The active controller will also check that any given broker stay fenced until it learns of all partition to log directory assignments in that specific broker via the new AssignReplicasToDirs RPC.
  • During the migration, replicas are assumed and assigned to log directory Uuid.ZERO until the actual log directory is learnt by the active controller from a broker running in KRaft mode.

Replica management

Existing replicas without a log directory are either:

  • Assumed to live in a broker that isn’t yet configured with multiple log directories, and so live in a single log directory, even if the UUID for that directory isn’t yet known by the controller. It is not possible to trigger a log directory failure from a broker that has a single log directory, as the broker would simply shut down if there are no remaining online log directories. Or
  • Assigned to a log directory as of yet unknown, in a broker that remains FENCED. As the broker remains FENCED it cannot assume leadership for any partition, and so a log directory failure would be handled by the current partition leader.

The two assumptions above eliminate the risk of having a broker which is not shutting down, but is unable to continue its leadership responsibilities due to the partition being persisted in a log directory that is broken or otherwise unavailable and the active controller not being aware of such an issue. 

Storage formatting

The changes to storage formatting simply ensure the existence of two new types of metadata files at the log directory roots. 

Test plan

The system test for log directory failures will be extended to KRaft mode.

Future work

  • Partition reassignment across directories and across brokers involves different API calls — AlterPartitionReassignments and AlterReplicaLogDirs. Whilst reassigning partitions across brokers into a specific log directory is already possible, it involves an intricate sequence of calls previous calls to AlterReplicaLogDirs and expecting errors as a successful result. Once this work is done we can consolidate these two API calls by extending AlterPartitionReassignments to allow target log directories to be specified and deprecate AlterReplicaLogDirs. This can be done as part of a future KIP.
  • The only way to know which log directory UUID corresponds to which log directory path is by reading the  files in each broker. A future KIP should expand the DescribeLogDirs RPC response to include log directory UUIDs along with the system path for each log directory.
  • Partition initialization can be optimized, by having the controller preselect a log directory for new partitions. This would avoid having to wait for the broker to send a AssignReplicasToDirs request to indicate the chosen log directory before it is safe for the broker to assume leadership of the partition. Maybe the controller could also take available storage in each log directory into account if the broker indicates the available storage space for each log directory as part of broker registration. This may be be proposed in a future KIP.

Rejected alternatives

  • Keeping the scope of the log directory to the broker — while this would mean a much simpler change, as was proposed in KIP-589, if only the broker itself knows which partitions were assigned to a log directory, when a log directory fails the broker will need to send a potentially very large request enumerating all the partitions in the failed disk, so that the controller can update leadership and ISRs accordingly.
  • Having the controller determine the log directory for new replicas — this would avoid a further RPC from the broker upon selecting a new log directory for new replicas, and reduce the time until it is safe for the broker to take leadership of the replica. However the broker is in a better position to make a choice of log directory than the controller, as it has easier access to e.g. disk usage in each log directory. The controller could also have this information if the broker were to include it the broker registration. But to keep the design simple, this optimization is best left for future work.
  • Changing how log directory failure is handled in ZooKeeper mode — ZooKeeper mode is going away, KIP-833 proposed its deprecation in a near future release.
  • Using the system path to identify each log directory, or storing the identifier somewhere else — When running Kafka with multiple log directories, each log directory is typically assigned to a different system disk or volume. The same storage device can be made accessible under a different mount, and Kafka should be able to identify the contents as the same disk. Because the log directory configuration can change on the broker, the only reliable way to identify the log directory in each broker is to add metadata to the file system under the log directory itself.
  • Fail on broker startup if the cluster metadata indicates replicas should be in a different log directory than the directory where the broker actually finds them — Despite not being an advertised feature, currently replicas can be moved between log directories while the broker is offline. Once the broker comes back up it accepts the new location of the replica. To continue supporting this feature, the broker will need to compare the information in the cluster metadata with the actual replica location during startup and take action on any mismatch.
  • Not keeping a list of all configured log directory identifiers in a metadata file in the file system — If a broker finds some log directories to be offline during startup it will need another way to identify the log directory as the metadata is in the disk itself. For this reason, the list of all log directory IDs must also be persisted in somewhere that the broker can expected to be able to access. Regardless of how many log directories are configured, one of the log directories will be configured to host the cluster metadata topic. This can either be one of the data log directories or a completely separate one — typically in the system disk. Since the broker will not startup if this log directory is not available, it is the perfect location to persist the list of all log directory identifiers.


  1. The exception is the cluster metadata log directory, which can be configured separately with metadata.log.dir . If the metadata log directory fails, then the broker cannot continue to run. If the broker isn't running it won't send any heartbeats and the controller will know to reassign leadership and update ISRs.The main benefit of support for multiple log directories is to allow brokers to continue operating if any single one of them fails. Typically, each log directory is mapped to an independent storage device while this critical metadata log directory would instead be mapped to the one of the main system partitions.

  • No labels