Status

Current state: Under DiscussionAccepted

Discussion thread: https://www.mail-archive.com/dev@kafka.apache.org/msg83115.html

...

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

Motivation

Apache Kafka Brokers make periodic FetchRequests to other brokers, in order to learn about updates to partitions they are following.  These periodic FetchRequests must enumerate all the partitions which the follower is interested in.  The responses also enumerate all the partitions, plus metadata (and potentially data) about each one.

...

We can solve the scalability and latency problems discussed above by creating "incremental" fetch requests and responses that only include information about what has changed.  In order to do this, we need to introduce the concept of "fetch sessions."

...

Because fetch sessions use memory on the leader, we want to limit the amount of them that we have at any given time.  Therefore, each broker will create only a limited number of incremental fetch sessions.

There are two is one new public configurations for fetch session caching:

• max.incremental.fetch.session.cache.slots, which set the number of incremental fetch session cache slots on the broker.  Default value: 1,000

There is one new constant for fetch session caching:

• min.incremental.fetch.session.eviction.ms, which sets the minimum amount of time we will wait before evicting an incremental fetch session from the cache.  Value: 120,000

When the server gets a new request to create an incremental fetch session, it will compare the proposed new session with its existing sessions.  The new session will evict an existing session if and only if:

...

There are several changes to the FetchRequest API.

Fetch Session ID

The fetch session ID uniquely A 32-bit number which identifies the current fetcherfetch session.  If this is set to 0, the server may create a new fetch session and return its value in the FetchResponse.

If this is set to a non-zero value, it is the fetch session ID.

Fetch Type

There is now a fetch_type field with the following values:

FULL_FETCH_REQUEST(0): This is a full fetch request.

INCREMENTAL_FETCH_REQUEST(1): This is an incremental fetch request.

The response to a full fetch request is always a full fetch response.  Similarly, the response to an incremental fetch request is always an incremental fetch response.

Incremental FetchRequests will only contain information about partitions which have changed on the follower.  If a partition has not changed, it will not be included. Possible changes include:

• Needing to fetch from a new offset in the partition
• Changing the maximum number of bytes which should be fetched from the partition
• Changes in the logStartOffset of the partition on the follower

Note that the existing system by which the leader becomes aware of changes to the follower's fetch position is not changed.  Every time the follower wants to update the fetch position, it must include the partition in the next incremental request.  Also note that if no partitions have changed, the next incremental FetchRequest will not contain any partitions at all.

An incremental fetch request cannot be made with a 0 fetch session ID.  An incremental fetch request should only be made if the server has returned INCREMENTAL_FETCH_OK in a previous full fetch request (see FetchResponse changes below)

Full FetchRequests and full FetchResponses will contain information about all partitions.  This is unchanged from the current behavior.  The client can alter the set of partitions in the fetch session by making a full fetch request with the designated session ID.

Incremental Fetch Sequence Number

The incremental fetch sequence number is a monotonically increasing counter. It is used for several purposes:

...

there is no current fetch session.

Fetch Session Epoch

A 32-bit number which identifies the current fetch session epoch.  Valid session epochs are always positive-- they are never 0 or negative numbers.

The fetch session epoch is incremented by one for each fetch request that we send.  Once it reaches MAX_INT, the next epoch is 1.

The fetch epoch keeps the state on the leader and the follower synchronized.  It ensures that if a message is duplicated or lost, the server will always notice.  It is also used to associate requests and responses in the logs. Other numbers, such as IP addresses and ports, or NetworkClient sequence numbers, can also be helpful for this purpose-- but they are less likely to be unique.

FetchRequest Metadata meaning

Incremental Fetch Requests

Incremental fetch requests have a positive fetch session ID.

A partition is only included in an incremental FetchRequest if:

• The client wants to notify the broker about a change to the partition's maxBytes, fetchOffset, or logStart
• The partition was not included in the incremental fetch session before, but the client wants to add it.
• The partition is in the incremental fetch session, but the client wants to remove it.

If the client doesn't want to change anything, the client does not need to include any partitions in the request at all.

If the client wants to remove a partition, the client will add the partition to the removed_partition list in the relevant removed_topics entry

...

.

Schema

FetchRequest => max_wait_time replica_id min_bytes isolation_level fetch_session_id fetch_session_epoch [topic] [removed_topic]

  max_wait_time => INT32

...

  min_bytes => INT32

  isolation_level => INT8  fetch_type => INT8

  fetch_session_id => INT64INT32

  fetch_session_sequence_number epoch => INT64INT32

  topic => topic_name [partition]

...

  partition => partition_id fetch_offset start_offset max_bytes

  partition_id => STRINGINT32

  fetch_offset => INT64

  start_offset => INT64

  max_bytes => INT32.

  removed_topic => removed_topic_name [removed_partition_id]

  removed_topic_name => STRING

  removed_partition_id => INT32

FetchResponse Changes

Top-level error code

Per-partition error codes are no longer sufficient to handle all response errors.  For example, when an incremental fetch session encounters an InvalidFetchSessionExceptionFetchSessionIdNotFoundException, we do not know which partitions the client expected to fetch.  Therefore, the FetchResponse now contains a top-level error code.  This error code is set to indicate that the request as a whole cannot be processed.

When the top-level error code is set, the caller should assume that all the partitions in the fetch received the given error.

Fetch Session ID

The server uses this field to notify the client when it has created a new FetchResponse now contains a 32-bit fetch session ID.

This field will be 0 if the server has closed the request session, or declined to create it in the first place.  Otherwise, it will be the session ID which the client should use for future requests.

Response Flags

response_flags is a bitfield.

INCREMENTAL_FETCH_RESPONSE(0): This bit is set if the response is an incremental fetch response.  It is cleared if the response is a full fetch response.

INCREMENTAL_FETCH_OK(1): This bit is set if the client can make incremental fetch requests with this session in the future.  It is cleared if the client should make full fetch requests in the future

Incremental Fetch Responses

Incremental fetch responses will only contain information about partitions for which:

• The logStartOffset has changed, or
• The highWaterMark has changed, or
• The lastStableOffset has changed, or
• The aborted transaction list has changed, or
• There is partition data available

The format of the partition data within FetchResponse is unchanged.

Handling Partition Size Limits in Incremental Fetch Responses

Sometimes, the per-fetch-request limit is too small to allow us to return information about every partition.  In those cases, we will limit the number of partitions that we return information about, to avoid exceeding the per-request maximum.  (As specified in KIP-74, the response will always return at least one message, though.)

If we always returned partition information in the same order, we might end up "starving" the partitions which came near the end of the order.  With full fetch requests, the client can rotate the order in which it requests partitions in order to avoid this problem.  However, incremental fetch requests need not explicitly specify all the partitions.  Indeed, an incremental fetch request may contain no partitions at all.

In order to solve the starvation problem, the server must rotate the order in which it returns partition information. We do this by lexicographically ordering the partitions, and then starting at the partition whose index is the request sequence number modulo the number of partitions.

For example, suppose that the fetch session has partitions P1, P2, and P3.  In that case, the fetch request with sequence number 1 will start at index 1 % 3 = 1.  Therefore it will return information about P2 first.  If any space is left over, we will return information about P3.  And if any space is left over at the end, we will return information about P1. Similarly, for the request with sequence number 2, the order will be P3, P1, P2.  For the request with sequence number 3, the order will be P1, P2, P3.  And so on.

Schema

FetchResponse => throttle_time_ms error_code error_string fetch_session_id [topic]

FetchResponse Metadata meaning

Incremental Fetch Responses

A partition is only included in an incremental FetchResponse if:

• The broker wants to notify the client about a change to the partition's highWatermark or broker logStartOffset
• There is new data available for a partition

If the broker has no new information to report, it does not need to include any partitions in the response at all.

The format of the partition data within FetchResponse is unchanged.

Handling Partition Size Limits in Incremental Fetch Responses

Sometimes, the per-fetch-request limit is too small to allow us to return information about every partition.  In those cases, we will limit the number of partitions that we return information about, to avoid exceeding the per-request maximum.  (As specified in KIP-74, the response will always return at least one message, though.)

If we always returned partition information in the same order, we might end up "starving" the partitions which came near the end of the order.  With full fetch requests, the client can rotate the order in which it requests partitions in order to avoid this problem.  However, incremental fetch requests need not explicitly specify all the partitions.  Indeed, an incremental fetch request may contain no partitions at all.

In order to solve the starvation problem, the server must rotate the order in which it returns partition information.  The server does this by maintaining a linked list of all partitions in the fetch session.  When data is returned for a partition, that partition is moved to the end of the list.  This ensures that we eventually return data about all partitions for which data is available.

Schema

FetchResponse => throttle_time_ms error_code error_string fetch_session_id [topic]

  throttle_time_ms => INT32

  error_code => INT16

  fetch_session_id => INT32

  topic => topic_name [partition]

  topic_name => STRING

  partition => partition_header records

  partition_header => partition_id error_code high_watermark last_stable_offset log_start_offset [aborted_transaction]

  partition_id   throttle_time_ms => INT32

  error_code => INT16

  errorhigh_string watermark => STRINGINT64

  fetchlast_sessionstable_id offset => INT64

  responselog_start_flags offset => INT8INT64

  topic aborted_transaction => topic_name [partition]producer_key first_offset

  topicproducer_name key => STRINGINT64

  partition first_offset => partition_header recordsINT64

  partition_header => partition_id error_code high_watermark last_stable_offset log_start_offset [aborted_transaction]

  partition_id => INT32

  error_code => INT16

  high_watermark => INT64

  last_stable_offset => INT64

  log_start_offset => INT64

  aborted_transaction => producer_key first_offset

  producer_key => INT64

  first_offset => INT64

  records => RECORDS

records => RECORDS

New Metrics

The following new metrics will be added to track cache consumption:

NumIncrementalFetchSessions: Tracks the number of incremental fetch sessions which exist.

NumIncrementalFetchPartitionsCached: Tracks the total number of partitions cached by incremental fetch sessions.

IncrementalFetchSessionEvictionsPerSec: Tracks the number of incremental fetch sessions that were evicted from the cache per second. This metric is not increased when a client closes its own incremental fetch session 

Compatibility, Deprecation, and Migration Plan

...