Web UI
The K2K-Tool Web UI allows management of committed offsets.
The K2K Tool provides a web interface that can be accessed when running the server command.
Interface
The web interface consists of:
Left Sidebar: Lists the managed pipelines. Each pipeline offers two views — K2K Record Replication and K2K Offset Mapping.
Centre Content Panel: The table for the selected view.
Information Sidebar (right): Per-partition detail for the selected topic or consumer group. Drag its left edge to resize it; the width is remembered for later sessions.
Data updates automatically over a WebSocket connection — there is no manual refresh. The first snapshot arrives on connect and later ones follow the refresh interval, bounded by --websocket-min-refresh-interval. If the connection drops, the pipeline list stays visible while the UI reconnects; Kafka failures are surfaced as an error in the UI.
Table updates can be paused while the connection stays open, which is useful when reading a value on a busy pipeline.
Views
K2K Record Replication
Shows how far the replicator has progressed for the pipeline's K2K consumer group. Topics expand inline to their partitions with committed, last, and lag columns. Topics and lag are sortable, and column headers carry help tooltips. The mapped target topic name is shown for each source topic-partition.

Lag is computed in the UI from the committed and last offsets. It is never shown as negative, and values come from cached reads, so they may be a few seconds behind.
K2K Offset Mapping
Available when offset mapping is enabled. Lists the consumer groups matched by the pipeline's groups.consumerGroups filter, one row per group, with Committed (source/target), Commit Lag (source/target), and Offset Mapping Lag (estimate).
Selecting a group opens the information sidebar, which breaks the same values down per partition and adds the source and target last offsets and the topic name on each side (the target name differs from the source when sink rename rules apply).
The Offset Mapping Lag (estimate) column shows 0 when an exact mapping exists for the source committed offset. Otherwise it estimates how far that commit is past the pipeline's last replicated source offset.
Separate filters are available for consumer group name and topic name.

Actions
Set offsets
Select a topic in Record Replication
Enter new committed offsets per partition in one dialog. Each partition reports its own success or failure.
Reset
Partition row
Reset that partition's committed offset.
Reset All
Above the committed offsets table
Reset all partitions of all topics in the pipeline to offset 0 (earliest).
Reset dialogs warn you to stop all replicator instances for the pipeline before continuing. If an instance is still running, it may overwrite the change.
Run
Pipelines and server
--file / -f
Comma-separated list of file(s) that define one k2k pipeline. Can be specified multiple times.
-
Yes, unless --dir is specified
--dir
List of directories to watch for pipeline files. Can be specified multiple times.
-
Yes, unless --file is specified
--defaults / -d
File containing defaults for the k2k app. Can be specified multiple times.
-
No
--host / -h
Host the server will accept requests from.
0.0.0.0
No
--port / -p
Port the server will accept requests from.
8080
No
--template-engine / -g
Enable or disable the template engine (enabled or disabled).
enabled
No
--verbosity
Sets the log verbosity. One of off, error, info, debug, trace.
-
No
Background refresh
The server keeps cached views of cluster health, topic end offsets, and committed consumer group offsets, refreshed in the background. Lengthen these intervals to reduce admin API load against the clusters — particularly on managed offerings that throttle admin requests — at the cost of slightly staler numbers in the UI.
--healthcheck-timeout / -e
Duration to wait for a health check to complete.
5s
No
--healthcheck-interval / -H
Duration between background health checks against the configured clusters.
30s
No
--topic-offset-interval / -O
Duration between background source topic end-offset refreshes for managed pipelines.
1.5s
No
--topic-offset-batch-size / -B
Maximum number of topic-partitions fetched per end-offset request batch.
100
No
--consumer-group-offset-interval / -G
Duration between background committed consumer group offset refreshes for offset-mapping pipelines.
5s
No
--kafka-fetch-jitter / -J
Fraction of the base interval added as random delay between Kafka fetch cycles. 0 disables; 0.2 adds up to 20%. Must be between 0 and 1.
0
No
--watcher-interval / -w
Duration between checks for new files in the managed directories.
10s
No
--websocket-min-refresh-interval / -R
Minimum duration between WebSocket snapshot updates. Client refresh values below this are raised to the minimum.
500ms
No
Use --kafka-fetch-jitter when running several k2k-tool servers against the same clusters, so their background refresh cycles do not align into bursts.
Transient Kafka errors do not stop background refresh: a failed fetch is logged and skipped so the rest of that cycle still runs, and health status resets to pending until the next successful check. Consumer groups and topic partitions that no longer exist are dropped from the live view.
Examples
HTTP API
The server command serves the UI and a JSON API on the same host and port.
GET
/
Web UI
GET
/api/health
Cached cluster health. 200 when healthy, 503 when not
GET
/api/pipelines
Names of the managed pipelines
GET
/api/groups?group=&pipeline=
Consumer group state, optionally filtered by group and pipeline
POST
/api/pipelines/{pipeline}/group/reset
Reset all partitions of all topics to offset 0
PUT
/api/pipelines/{pipeline}/group
Set specific committed offsets (JSON body)
GET
/api/ws?refresh=<duration>
WebSocket stream of live snapshots
/api/health is suitable as a Kubernetes liveness or readiness probe, since it returns 503 while any configured cluster is unreachable:
The reset and set endpoints change committed offsets on the target Kafka cluster — the same warning about stopping replicator instances first applies.
WebSocket stream
A client subscribes to one data stream at a time and receives snapshots for that type only:
subscribeGroups
K2K consumer group offsets (record replication)
subscribeOffsetMapping
offsetMappingGroups — offset-mapping consumer group state
The first snapshot is sent immediately on connect. Later snapshots follow the refresh query parameter, which defaults to --websocket-min-refresh-interval when the client omits it and is raised to that minimum when the client requests something shorter. Unknown subscription operations are logged and ignored rather than closing the connection.
Last updated
Was this helpful?

