The Storm UI daemon provides a REST API that allows you to interact with a Storm cluster, which includes retrieving metrics data and configuration information as well as management operations such as starting or stopping topologies.
The REST API returns JSON responses and supports JSONP. Clients can pass a callback query parameter to wrap JSON in the callback function.
Note: It is recommended to ignore undocumented elements in the JSON response because future versions of Storm may not support those elements anymore.
The REST API is part of the UI daemon of Storm (started by storm ui
) and thus runs on the same host and port as the
Storm UI (the UI daemon is often run on the same host as the Nimbus daemon). The port is configured by ui.port
,
which is set to 8080
by default (see defaults.yaml).
The API base URL would thus be:
http://<ui-host>:<ui-port>/api/v1/...
You can use a tool such as curl
to talk to the REST API:
# Request the cluster configuration.
# Note: We assume ui.port is configured to the default value of 8080.
$ curl http://<ui-host>:8080/api/v1/cluster/configuration
In a secure environment an authenticated user can impersonate another user. To impersonate a user the caller must pass
doAsUser
param or header with value set to the user that the request needs to be performed as. Please see SECURITY.MD
to learn more about how to setup impersonation ACLs and authorization. The rest API uses the same configs and acls that
are used by nimbus.
Examples:
1. http://ui-daemon-host-name:8080/api/v1/topology/wordcount-1-1425844354\?doAsUser=testUSer1
2. curl 'http://localhost:8080/api/v1/topology/wordcount-1-1425844354/activate' -X POST -H 'doAsUser:testUSer1'
Returns the cluster configuration.
Sample response (does not include all the data fields):
{
"dev.zookeeper.path": "/tmp/dev-storm-zookeeper",
"topology.tick.tuple.freq.secs": null,
"topology.builtin.metrics.bucket.size.secs": 60,
"topology.fall.back.on.java.serialization": true,
"topology.max.error.report.per.interval": 5,
"zmq.linger.millis": 5000,
"topology.skip.missing.kryo.registrations": false,
"storm.messaging.netty.client_worker_threads": 1,
"ui.childopts": "-Xmx768m",
"storm.zookeeper.session.timeout": 20000,
"nimbus.reassign": true,
"topology.trident.batch.emit.interval.millis": 500,
"storm.messaging.netty.flush.check.interval.ms": 10,
"nimbus.monitor.freq.secs": 10,
"logviewer.childopts": "-Xmx128m",
"java.library.path": "/usr/local/lib:/opt/local/lib:/usr/lib",
"topology.executor.send.buffer.size": 1024,
}
Returns cluster summary information such as nimbus uptime or number of supervisors.
Response fields:
Field | Value | Description |
---|---|---|
stormVersion | String | Storm version |
supervisors | Integer | Number of supervisors running |
topologies | Integer | Number of topologies running |
slotsTotal | Integer | Total number of available worker slots |
slotsUsed | Integer | Number of worker slots used |
slotsFree | Integer | Number of worker slots available |
executorsTotal | Integer | Total number of executors |
tasksTotal | Integer | Total tasks |
schedulerDisplayResource | Boolean | Whether to display scheduler resource information |
totalMem | Double | The total amount of memory in the cluster in MB |
totalCpu | Double | The total amount of CPU in the cluster |
availMem | Double | The amount of available memory in the cluster in MB |
availCpu | Double | The amount of available cpu in the cluster |
memAssignedPercentUtil | Double | The percent utilization of assigned memory resources in cluster |
cpuAssignedPercentUtil | Double | The percent utilization of assigned CPU resources in cluster |
Sample response:
{
"stormVersion": "0.9.2-incubating-SNAPSHOT",
"supervisors": 1,
"slotsTotal": 4,
"slotsUsed": 3,
"slotsFree": 1,
"executorsTotal": 28,
"tasksTotal": 28,
"schedulerDisplayResource": true,
"totalMem": 4096.0,
"totalCpu": 400.0,
"availMem": 1024.0,
"availCPU": 250.0,
"memAssignedPercentUtil": 75.0,
"cpuAssignedPercentUtil": 37.5
}
Returns summary information for all supervisors.
Response fields:
Field | Value | Description |
---|---|---|
id | String | Supervisor's id |
host | String | Supervisor's host name |
uptime | String | Shows how long the supervisor is running |
uptimeSeconds | Integer | Shows how long the supervisor is running in seconds |
slotsTotal | Integer | Total number of available worker slots for this supervisor |
slotsUsed | Integer | Number of worker slots used on this supervisor |
schedulerDisplayResource | Boolean | Whether to display scheduler resource information |
totalMem | Double | Total memory capacity on this supervisor |
totalCpu | Double | Total CPU capacity on this supervisor |
usedMem | Double | Used memory capacity on this supervisor |
usedCpu | Double | Used CPU capacity on this supervisor |
Sample response:
{
"supervisors": [
{
"id": "0b879808-2a26-442b-8f7d-23101e0c3696",
"host": "10.11.1.7",
"uptime": "5m 58s",
"uptimeSeconds": 358,
"slotsTotal": 4,
"slotsUsed": 3,
"totalMem": 3000,
"totalCpu": 400,
"usedMem": 1280,
"usedCPU": 160
}
],
"schedulerDisplayResource": true
}
Returns summary information for all nimbus hosts.
Response fields:
Field | Value | Description |
---|---|---|
host | String | Nimbus' host name |
port | int | Nimbus' port number |
status | String | Possible values are Leader, Not a Leader, Dead |
nimbusUpTime | String | Shows since how long the nimbus has been running |
nimbusUpTimeSeconds | String | Shows since how long the nimbus has been running in seconds |
nimbusLogLink | String | Logviewer url to view the nimbus.log |
version | String | Version of storm this nimbus host is running |
Sample response:
{
"nimbuses":[
{
"host":"192.168.202.1",
"port":6627,
"nimbusLogLink":"http:\/\/192.168.202.1:8000\/log?file=nimbus.log",
"status":Leader,
"version":"0.10.0-SNAPSHOT",
"nimbusUpTime":"3m 33s",
"nimbusUpTimeSeconds":"213"
}
]
}
Returns a list of all running topologies' IDs submitted by the current user.
Response fields:
Field | Value | Description |
---|---|---|
topo-history | List | List of Topologies' IDs |
Sample response:
{
"topo-history":[
"wc6-1-1446571009",
"wc8-2-1446587178"
]
}
Returns summary for a supervisor by id, or all supervisors running on a host.
Examples:
1. By host: http://ui-daemon-host-name:8080/api/v1/supervisor?host=supervisor-daemon-host-name
2. By id: http://ui-daemon-host-name:8080/api/v1/supervisor?id=f5449110-1daa-43e2-89e3-69917b16dec9-192.168.1.1
Request parameters:
Parameter | Value | Description |
---|---|---|
id | String. Supervisor id | If specified, respond with the supervisor and worker stats with id. Note that when id is specified, the host argument is ignored. |
host | String. Host name | If specified, respond with all supervisors and worker stats in the host (normally just one) |
sys | String. Values 1 or 0. Default value 0 | Controls including sys stats part of the response |
Response fields:
Field | Value | Description |
---|---|---|
supervisors | Array | Array of supervisor summaries |
workers | Array | Array of worker summaries |
schedulerDisplayResource | Boolean | Whether to display scheduler resource information |
Each supervisor is defined by:
Field | Value | Description |
---|---|---|
id | String | Supervisor's id |
host | String | Supervisor's host name |
uptime | String | Shows how long the supervisor is running |
uptimeSeconds | Integer | Shows how long the supervisor is running in seconds |
slotsTotal | Integer | Total number of worker slots for this supervisor |
slotsUsed | Integer | Number of worker slots used on this supervisor |
schedulerDisplayResource | Boolean | Whether to display scheduler resource information |
totalMem | Double | Total memory capacity on this supervisor |
totalCpu | Double | Total CPU capacity on this supervisor |
usedMem | Double | Used memory capacity on this supervisor |
usedCpu | Double | Used CPU capacity on this supervisor |
Each worker is defined by:
Field | Value | Description |
---|---|---|
supervisorId | String | Supervisor's id |
host | String | Worker's host name |
port | Integer | Worker's port |
topologyId | String | Topology Id |
topologyName | String | Topology Name |
executorsTotal | Integer | Number of executors used by the topology in this worker |
assignedMemOnHeap | Double | Assigned On-Heap Memory by Scheduler (MB) |
assignedMemOffHeap | Double | Assigned Off-Heap Memory by Scheduler (MB) |
assignedCpu | Number | Assigned CPU by Scheduler (%) |
componentNumTasks | Dictionary | Components -> # of executing tasks |
uptime | String | Shows how long the worker is running |
uptimeSeconds | Integer | Shows how long the worker is running in seconds |
workerLogLink | String | Link to worker log viewer page |
Sample response:
{
"supervisors": [{
"totalMem": 4096.0,
"host":"192.168.10.237",
"id":"bdfe8eff-f1d8-4bce-81f5-9d3ae1bf432e-169.254.129.212",
"uptime":"7m 8s",
"totalCpu":400.0,
"usedCpu":495.0,
"usedMem":3432.0,
"slotsUsed":2,
"version":"0.10.1",
"slotsTotal":4,
"uptimeSeconds":428
}],
"schedulerDisplayResource":true,
"workers":[{
"topologyName":"ras",
"topologyId":"ras-4-1460229987",
"host":"192.168.10.237",
"supervisorId":"bdfe8eff-f1d8-4bce-81f5-9d3ae1bf432e-169.254.129.212",
"assignedMemOnHeap":704.0,
"uptime":"2m 47s",
"uptimeSeconds":167,
"port":6707,
"workerLogLink":"http:\/\/192.168.10.237:8000\/log?file=ras-4-1460229987%2F6707%2Fworker.log",
"componentNumTasks": {
"word":5
},
"executorsTotal":8,
"assignedCpu":130.0,
"assignedMemOffHeap":80.0
},
{
"topologyName":"ras",
"topologyId":"ras-4-1460229987",
"host":"192.168.10.237",
"supervisorId":"bdfe8eff-f1d8-4bce-81f5-9d3ae1bf432e-169.254.129.212",
"assignedMemOnHeap":904.0,
"uptime":"2m 53s",
"port":6706,
"workerLogLink":"http:\/\/192.168.10.237:8000\/log?file=ras-4-1460229987%2F6706%2Fworker.log",
"componentNumTasks":{
"exclaim2":2,
"exclaim1":3,
"word":5
},
"executorsTotal":10,
"uptimeSeconds":173,
"assignedCpu":165.0,
"assignedMemOffHeap":80.0
}]
}
Returns summary information for all topologies.
Response fields:
Field | Value | Description |
---|---|---|
id | String | Topology Id |
name | String | Topology Name |
status | String | Topology Status |
uptime | String | Shows how long the topology is running |
uptimeSeconds | Integer | Shows how long the topology is running in seconds |
tasksTotal | Integer | Total number of tasks for this topology |
workersTotal | Integer | Number of workers used for this topology |
executorsTotal | Integer | Number of executors used for this topology |
replicationCount | Integer | Number of nimbus hosts on which this topology code is replicated |
requestedMemOnHeap | Double | Requested On-Heap Memory by User (MB) |
requestedMemOffHeap | Double | Requested Off-Heap Memory by User (MB) |
requestedTotalMem | Double | Requested Total Memory by User (MB) |
requestedCpu | Double | Requested CPU by User (%) |
assignedMemOnHeap | Double | Assigned On-Heap Memory by Scheduler (MB) |
assignedMemOffHeap | Double | Assigned Off-Heap Memory by Scheduler (MB) |
assignedTotalMem | Double | Assigned Total Memory by Scheduler (MB) |
assignedCpu | Double | Assigned CPU by Scheduler (%) |
schedulerDisplayResource | Boolean | Whether to display scheduler resource information |
Sample response:
{
"topologies": [
{
"id": "WordCount3-1-1402960825",
"name": "WordCount3",
"status": "ACTIVE",
"uptime": "6m 5s",
"uptimeSeconds": 365,
"tasksTotal": 28,
"workersTotal": 3,
"executorsTotal": 28,
"replicationCount": 1,
"requestedMemOnHeap": 640,
"requestedMemOffHeap": 128,
"requestedTotalMem": 768,
"requestedCpu": 80,
"assignedMemOnHeap": 640,
"assignedMemOffHeap": 128,
"assignedTotalMem": 768,
"assignedCpu": 80
}
],
"schedulerDisplayResource": true
}
Returns the worker' information (host and port) for a topology.
Response fields:
Field | Value | Description |
---|---|---|
hostPortList | List | Workers' information for a topology |
name | Integer | Logviewer Port |
Sample response:
{
"hostPortList":[
{
"host":"192.168.202.2",
"port":6701
},
{
"host":"192.168.202.2",
"port":6702
},
{
"host":"192.168.202.3",
"port":6700
}
],
"logviewerPort":8000
}
Returns topology information and statistics. Substitute id with topology id.
Request parameters:
Parameter | Value | Description |
---|---|---|
id | String (required) | Topology Id |
window | String. Default value :all-time | Window duration for metrics in seconds |
sys | String. Values 1 or 0. Default value 0 | Controls including sys stats part of the response |
Response fields:
Field | Value | Description |
---|---|---|
id | String | Topology Id |
name | String | Topology Name |
uptime | String | How long the topology has been running |
uptimeSeconds | Integer | How long the topology has been running in seconds |
status | String | Current status of the topology, e.g. "ACTIVE" |
tasksTotal | Integer | Total number of tasks for this topology |
workersTotal | Integer | Number of workers used for this topology |
executorsTotal | Integer | Number of executors used for this topology |
msgTimeout | Integer | Number of seconds a tuple has before the spout considers it failed |
windowHint | String | window param value in "hh mm ss" format. Default value is "All Time" |
schedulerDisplayResource | Boolean | Whether to display scheduler resource information |
replicationCount | Integer | Number of nimbus hosts on which this topology code is replicated |
debug | Boolean | If debug is enabled for the topology |
samplingPct | Double | Controls downsampling of events before they are sent to event log (percentage) |
assignedMemOnHeap | Double | Assigned On-Heap Memory by Scheduler (MB) |
assignedMemOffHeap | Double | Assigned Off-Heap Memory by Scheduler (MB) |
assignedTotalMem | Double | Assigned Off-Heap + On-Heap Memory by Scheduler(MB) |
assignedCpu | Double | Assigned CPU by Scheduler(%) |
requestedMemOnHeap | Double | Requested On-Heap Memory by User (MB) |
requestedMemOffHeap | Double | Requested Off-Heap Memory by User (MB) |
requestedCpu | Double | Requested CPU by User (%) |
topologyStats | Array | Array of all the topology related stats per time window |
topologyStats.windowPretty | String | Duration passed in HH:MM:SS format |
topologyStats.window | String | User requested time window for metrics |
topologyStats.emitted | Long | Number of messages emitted in given window |
topologyStats.trasferred | Long | Number messages transferred in given window |
topologyStats.completeLatency | String (double value returned in String format) | Total latency for processing the message |
topologyStats.acked | Long | Number of messages acked in given window |
topologyStats.failed | Long | Number of messages failed in given window |
workers | Array | Array of workers in topology |
workers.supervisorId | String | Supervisor's id |
workers.host | String | Worker's host name |
workers.port | Integer | Worker's port |
workers.topologyId | String | Topology Id |
workers.topologyName | String | Topology Name |
workers.executorsTotal | Integer | Number of executors used by the topology in this worker |
workers.assignedMemOnHeap | Double | Assigned On-Heap Memory by Scheduler (MB) |
workers.assignedMemOffHeap | Double | Assigned Off-Heap Memory by Scheduler (MB) |
workers.assignedCpu | Number | Assigned CPU by Scheduler (%) |
workers.componentNumTasks | Dictionary | Components -> # of executing tasks |
workers.uptime | String | Shows how long the worker is running |
workers.uptimeSeconds | Integer | Shows how long the worker is running in seconds |
workers.workerLogLink | String | Link to worker log viewer page |
spouts | Array | Array of all the spout components in the topology |
spouts.spoutId | String | Spout id |
spouts.executors | Integer | Number of executors for the spout |
spouts.emitted | Long | Number of messages emitted in given window |
spouts.completeLatency | String (double value returned in String format) | Total latency for processing the message |
spouts.transferred | Long | Total number of messages transferred in given window |
spouts.tasks | Integer | Total number of tasks for the spout |
spouts.lastError | String | Shows the last error happened in a spout |
spouts.errorLapsedSecs | Integer | Number of seconds elapsed since that last error happened in a spout |
spouts.errorWorkerLogLink | String | Link to the worker log that reported the exception |
spouts.acked | Long | Number of messages acked |
spouts.failed | Long | Number of messages failed |
spouts.requestedMemOnHeap | Double | Requested On-Heap Memory by User (MB) |
spouts.requestedMemOffHeap | Double | Requested Off-Heap Memory by User (MB) |
spouts.requestedCpu | Double | Requested CPU by User (%) |
bolts | Array | Array of bolt components in the topology |
bolts.boltId | String | Bolt id |
bolts.capacity | String (double value returned in String format) | This value indicates number of messages executed * average execute latency / time window |
bolts.processLatency | String (double value returned in String format) | Average time of the bolt to ack a message after it was received |
bolts.executeLatency | String (double value returned in String format) | Average time to run the execute method of the bolt |
bolts.executors | Integer | Number of executor tasks in the bolt component |
bolts.tasks | Integer | Number of instances of bolt |
bolts.acked | Long | Number of tuples acked by the bolt |
bolts.failed | Long | Number of tuples failed by the bolt |
bolts.lastError | String | Shows the last error occurred in the bolt |
bolts.errorLapsedSecs | Integer | Number of seconds elapsed since that last error happened in a bolt |
bolts.errorWorkerLogLink | String | Link to the worker log that reported the exception |
bolts.emitted | Long | Number of tuples emitted |
bolts.requestedMemOnHeap | Double | Requested On-Heap Memory by User (MB) |
bolts.requestedMemOffHeap | Double | Requested Off-Heap Memory by User (MB) |
bolts.requestedCpu | Double | Requested CPU by User (%) |
Examples:
1. http://ui-daemon-host-name:8080/api/v1/topology/WordCount3-1-1402960825
2. http://ui-daemon-host-name:8080/api/v1/topology/WordCount3-1-1402960825?sys=1
3. http://ui-daemon-host-name:8080/api/v1/topology/WordCount3-1-1402960825?window=600
Sample response:
{
"name": "WordCount3",
"id": "WordCount3-1-1402960825",
"workersTotal": 3,
"window": "600",
"status": "ACTIVE",
"tasksTotal": 28,
"executorsTotal": 28,
"uptime": "29m 19s",
"uptimeSeconds": 1759,
"msgTimeout": 30,
"windowHint": "10m 0s",
"schedulerDisplayResource": true,
"topologyStats": [
{
"windowPretty": "10m 0s",
"window": "600",
"emitted": 397960,
"transferred": 213380,
"completeLatency": "0.000",
"acked": 213460,
"failed": 0
},
{
"windowPretty": "3h 0m 0s",
"window": "10800",
"emitted": 1190260,
"transferred": 638260,
"completeLatency": "0.000",
"acked": 638280,
"failed": 0
},
{
"windowPretty": "1d 0h 0m 0s",
"window": "86400",
"emitted": 1190260,
"transferred": 638260,
"completeLatency": "0.000",
"acked": 638280,
"failed": 0
},
{
"windowPretty": "All time",
"window": ":all-time",
"emitted": 1190260,
"transferred": 638260,
"completeLatency": "0.000",
"acked": 638280,
"failed": 0
}
],
"workers":[{
"topologyName":"WordCount3",
"topologyId":"WordCount3-1-1402960825",
"host":"192.168.10.237",
"supervisorId":"bdfe8eff-f1d8-4bce-81f5-9d3ae1bf432e-169.254.129.212",
"uptime":"2m 47s",
"uptimeSeconds":167,
"port":6707,
"workerLogLink":"http:\/\/192.168.10.237:8000\/log?file=WordCount3-1-1402960825%2F6707%2Fworker.log",
"componentNumTasks": {
"spout":5
},
"executorsTotal":8,
"assignedMemOnHeap":704.0,
"assignedCpu":130.0,
"assignedMemOffHeap":80.0
}],
"spouts": [
{
"executors": 5,
"emitted": 28880,
"completeLatency": "0.000",
"transferred": 28880,
"acked": 0,
"spoutId": "spout",
"tasks": 5,
"lastError": "",
"errorLapsedSecs": null,
"failed": 0
}
],
"bolts": [
{
"executors": 12,
"emitted": 184580,
"transferred": 0,
"acked": 184640,
"executeLatency": "0.048",
"tasks": 12,
"executed": 184620,
"processLatency": "0.043",
"boltId": "count",
"lastError": "",
"errorLapsedSecs": null,
"capacity": "0.003",
"failed": 0
},
{
"executors": 8,
"emitted": 184500,
"transferred": 184500,
"acked": 28820,
"executeLatency": "0.024",
"tasks": 8,
"executed": 28780,
"processLatency": "2.112",
"boltId": "split",
"lastError": "",
"errorLapsedSecs": null,
"capacity": "0.000",
"failed": 0
}
],
"configuration": {
"storm.id": "WordCount3-1-1402960825",
"dev.zookeeper.path": "/tmp/dev-storm-zookeeper",
"topology.tick.tuple.freq.secs": null,
"topology.builtin.metrics.bucket.size.secs": 60,
"topology.fall.back.on.java.serialization": true,
"topology.max.error.report.per.interval": 5,
"zmq.linger.millis": 5000,
"topology.skip.missing.kryo.registrations": false,
"storm.messaging.netty.client_worker_threads": 1,
"ui.childopts": "-Xmx768m",
"storm.zookeeper.session.timeout": 20000,
"nimbus.reassign": true,
"topology.trident.batch.emit.interval.millis": 500,
"storm.messaging.netty.flush.check.interval.ms": 10,
"nimbus.monitor.freq.secs": 10,
"logviewer.childopts": "-Xmx128m",
"java.library.path": "/usr/local/lib:/opt/local/lib:/usr/lib",
"topology.executor.send.buffer.size": 1024,
"storm.local.dir": "storm-local",
"storm.messaging.netty.buffer_size": 5242880,
"supervisor.worker.start.timeout.secs": 120,
"topology.enable.message.timeouts": true,
"nimbus.cleanup.inbox.freq.secs": 600,
"nimbus.inbox.jar.expiration.secs": 3600,
"drpc.worker.threads": 64,
"topology.worker.shared.thread.pool.size": 4,
"nimbus.seeds": [
"hw10843.local"
],
"storm.messaging.netty.min_wait_ms": 100,
"storm.zookeeper.port": 2181,
"transactional.zookeeper.port": null,
"topology.executor.receive.buffer.size": 1024,
"transactional.zookeeper.servers": null,
"storm.zookeeper.root": "/storm",
"storm.zookeeper.retry.intervalceiling.millis": 30000,
"supervisor.enable": true,
"storm.messaging.netty.server_worker_threads": 1
},
"replicationCount": 1
}
Returns detailed metrics for topology. It shows metrics per component, which are aggregated by stream.
Parameter | Value | Description |
---|---|---|
id | String (required) | Topology Id |
window | String. Default value :all-time | window duration for metrics in seconds |
sys | String. Values 1 or 0. Default value 0 | Controls including sys stats part of the response |
Response fields:
Field | Value | Description |
---|---|---|
window | String. Default value ":all-time" | window duration for metrics in seconds |
windowHint | String | |
spouts | Array | Array of all the spout components in the topology |
spouts.id | String | Spout id |
spouts.emitted | Array | Array of all the output streams this spout emits messages |
spouts.emitted.stream_id | String | Stream id for this stream |
spouts.emitted.value | Long | Number of messages emitted in given window |
spouts.transferred | Array | Array of all the output streams this spout transfers messages |
spouts.transferred.stream_id | String | Stream id for this stream |
spouts.transferred.value | Long | Number messages transferred in given window |
spouts.acked | Array | Array of all the output streams this spout receives ack of messages |
spouts.acked.stream_id | String | Stream id for this stream |
spouts.acked.value | Long | Number of messages acked in given window |
spouts.failed | Array | Array of all the output streams this spout receives fail of messages |
spouts.failed.stream_id | String | Stream id for this stream |
spouts.failed.value | Long | Number of messages failed in given window |
spouts.complete_ms_avg | Array | Array of all the output streams this spout receives ack of messages |
spouts.complete_ms_avg.stream_id | String | Stream id for this stream |
spouts.complete_ms_avg.value | String (double value returned in String format) | Total latency for processing the message |
bolts | Array | Array of all the bolt components in the topology |
bolts.id | String | Bolt id |
bolts.emitted | Array | Array of all the output streams this bolt emits messages |
bolts.emitted.stream_id | String | Stream id for this stream |
bolts.emitted.value | Long | Number of messages emitted in given window |
bolts.transferred | Array | Array of all the output streams this bolt transfers messages |
bolts.transferred.stream_id | String | Stream id for this stream |
bolts.transferred.value | Long | Number messages transferred in given window |
bolts.acked | Array | Array of all the input streams this bolt acknowledges of messages |
bolts.acked.component_id | String | Component id for this stream |
bolts.acked.stream_id | String | Stream id for this stream |
bolts.acked.value | Long | Number of messages acked in given window |
bolts.failed | Array | Array of all the input streams this bolt receives fail of messages |
bolts.failed.component_id | String | Component id for this stream |
bolts.failed.stream_id | String | Stream id for this stream |
bolts.failed.value | Long | Number of messages failed in given window |
bolts.process_ms_avg | Array | Array of all the input streams this spout acks messages |
bolts.process_ms_avg.component_id | String | Component id for this stream |
bolts.process_ms_avg.stream_id | String | Stream id for this stream |
bolts.process_ms_avg.value | String (double value returned in String format) | Average time of the bolt to ack a message after it was received |
bolts.executed | Array | Array of all the input streams this bolt executes messages |
bolts.executed.component_id | String | Component id for this stream |
bolts.executed.stream_id | String | Stream id for this stream |
bolts.executed.value | Long | Number of messages executed in given window |
bolts.executed_ms_avg | Array | Array of all the output streams this spout receives ack of messages |
bolts.executed_ms_avg.component_id | String | Component id for this stream |
bolts.executed_ms_avg.stream_id | String | Stream id for this stream |
bolts.executed_ms_avg.value | String (double value returned in String format) | Average time to run the execute method of the bolt |
Examples:
1. http://ui-daemon-host-name:8080/api/v1/topology/WordCount3-1-1402960825/metrics
1. http://ui-daemon-host-name:8080/api/v1/topology/WordCount3-1-1402960825/metrics?sys=1
2. http://ui-daemon-host-name:8080/api/v1/topology/WordCount3-1-1402960825/metrics?window=600
Sample response:
{
"window":":all-time",
"window-hint":"All time",
"spouts":[
{
"id":"spout",
"emitted":[
{
"stream_id":"__metrics",
"value":20
},
{
"stream_id":"default",
"value":17350280
},
{
"stream_id":"__ack_init",
"value":17328160
},
{
"stream_id":"__system",
"value":20
}
],
"transferred":[
{
"stream_id":"__metrics",
"value":20
},
{
"stream_id":"default",
"value":17350280
},
{
"stream_id":"__ack_init",
"value":17328160
},
{
"stream_id":"__system",
"value":0
}
],
"acked":[
{
"stream_id":"default",
"value":17339180
}
],
"failed":[
],
"complete_ms_avg":[
{
"stream_id":"default",
"value":"920.497"
}
]
}
],
"bolts":[
{
"id":"count",
"emitted":[
{
"stream_id":"__metrics",
"value":120
},
{
"stream_id":"default",
"value":190748180
},
{
"stream_id":"__ack_ack",
"value":190718100
},
{
"stream_id":"__system",
"value":20
}
],
"transferred":[
{
"stream_id":"__metrics",
"value":120
},
{
"stream_id":"default",
"value":0
},
{
"stream_id":"__ack_ack",
"value":190718100
},
{
"stream_id":"__system",
"value":0
}
],
"acked":[
{
"component_id":"split",
"stream_id":"default",
"value":190733160
}
],
"failed":[
],
"process_ms_avg":[
{
"component_id":"split",
"stream_id":"default",
"value":"0.004"
}
],
"executed":[
{
"component_id":"split",
"stream_id":"default",
"value":190733140
}
],
"executed_ms_avg":[
{
"component_id":"split",
"stream_id":"default",
"value":"0.005"
}
]
},
{
"id":"split",
"emitted":[
{
"stream_id":"__metrics",
"value":60
},
{
"stream_id":"default",
"value":190754740
},
{
"stream_id":"__ack_ack",
"value":17317580
},
{
"stream_id":"__system",
"value":20
}
],
"transferred":[
{
"stream_id":"__metrics",
"value":60
},
{
"stream_id":"default",
"value":190754740
},
{
"stream_id":"__ack_ack",
"value":17317580
},
{
"stream_id":"__system",
"value":0
}
],
"acked":[
{
"component_id":"spout",
"stream_id":"default",
"value":17339180
}
],
"failed":[
],
"process_ms_avg":[
{
"component_id":"spout",
"stream_id":"default",
"value":"0.051"
}
],
"executed":[
{
"component_id":"spout",
"stream_id":"default",
"value":17339240
}
],
"executed_ms_avg":[
{
"component_id":"spout",
"stream_id":"default",
"value":"0.052"
}
]
}
]
}
Returns detailed metrics and executor information
Parameter | Value | Description |
---|---|---|
id | String (required) | Topology Id |
component | String (required) | Component Id |
window | String. Default value :all-time | window duration for metrics in seconds |
sys | String. Values 1 or 0. Default value 0 | controls including sys stats part of the response |
Response fields:
Field | Value | Description |
---|---|---|
user | String | Topology owner |
id | String | Component id |
encodedId | String | URL encoded component id |
name | String | Topology name |
executors | Integer | Number of executor tasks in the component |
tasks | Integer | Number of instances of component |
requestedMemOnHeap | Double | Requested On-Heap Memory by User (MB) |
requestedMemOffHeap | Double | Requested Off-Heap Memory by User (MB) |
requestedCpu | Double | Requested CPU by User (%) |
schedulerDisplayResource | Boolean | Whether to display scheduler resource information |
topologyId | String | Topology id |
topologyStatus | String | Topology status |
encodedTopologyId | String | URL encoded topology id |
window | String. Default value "All Time" | window duration for metrics in seconds |
componentType | String | component type: SPOUT or BOLT |
windowHint | String | window param value in "hh mm ss" format. Default value is "All Time" |
debug | Boolean | If debug is enabled for the component |
samplingPct | Double | Controls downsampling of events before they are sent to event log (percentage) |
eventLogLink | String | URL viewer link to event log (debug mode) |
profilingAndDebuggingCapable | Boolean | true if there is support for Profiling and Debugging Actions |
profileActionEnabled | Boolean | true if worker profiling (Java Flight Recorder) is enabled |
profilerActive | Array | Array of currently active Profiler Actions |
componentErrors | Array of Errors | List of component errors |
componentErrors.errorTime | Long | Timestamp when the exception occurred (Prior to 0.11.0, this field was named 'time'.) |
componentErrors.errorHost | String | host name for the error |
componentErrors.errorPort | String | port for the error |
componentErrors.error | String | Shows the error happened in a component |
componentErrors.errorLapsedSecs | Integer | Number of seconds elapsed since the error happened in a component |
componentErrors.errorWorkerLogLink | String | Link to the worker log that reported the exception |
spoutSummary | Array | (only for spouts) Array of component stats, one element per window. |
spoutSummary.windowPretty | String | Duration passed in HH:MM:SS format |
spoutSummary.window | String | window duration for metrics in seconds |
spoutSummary.emitted | Long | Number of messages emitted in given window |
spoutSummary.completeLatency | String (double value returned in String format) | Total latency for processing the message |
spoutSummary.transferred | Long | Total number of messages transferred in given window |
spoutSummary.acked | Long | Number of messages acked |
spoutSummary.failed | Long | Number of messages failed |
boltStats | Array | (only for bolts) Array of component stats, one element per window. |
boltStats.windowPretty | String | Duration passed in HH:MM:SS format |
boltStats.window | String | window duration for metrics in seconds |
boltStats.transferred | Long | Total number of messages transferred in given window |
boltStats.processLatency | String (double value returned in String format) | Average time of the bolt to ack a message after it was received |
boltStats.acked | Long | Number of messages acked |
boltStats.failed | Long | Number of messages failed |
inputStats | Array | (only for bolts) Array of input stats |
inputStats.component | String | Component id |
inputStats.encodedComponentId | String | URL encoded component id |
inputStats.executeLatency | Long | The average time a tuple spends in the execute method |
inputStats.processLatency | Long | The average time it takes to ack a tuple after it is first received |
inputStats.executed | Long | The number of incoming tuples processed |
inputStats.acked | Long | Number of messages acked |
inputStats.failed | Long | Number of messages failed |
inputStats.stream | String | The name of the tuple stream given in the topology, or "default" if none specified |
outputStats | Array | Array of output stats |
outputStats.transferred | Long | Number of tuples emitted that sent to one ore more bolts |
outputStats.emitted | Long | Number of tuples emitted |
outputStats.stream | String | The name of the tuple stream given in the topology, or "default" if none specified |
Examples:
1. http://ui-daemon-host-name:8080/api/v1/topology/WordCount3-1-1402960825/component/spout
2. http://ui-daemon-host-name:8080/api/v1/topology/WordCount3-1-1402960825/component/spout?sys=1
3. http://ui-daemon-host-name:8080/api/v1/topology/WordCount3-1-1402960825/component/spout?window=600
Sample response:
{
"name": "WordCount3",
"id": "spout",
"componentType": "spout",
"windowHint": "10m 0s",
"executors": 5,
"componentErrors":[{"errorTime": 1406006074000,
"errorHost": "10.11.1.70",
"errorPort": 6701,
"errorWorkerLogLink": "http://10.11.1.7:8000/log?file=worker-6701.log",
"errorLapsedSecs": 16,
"error": "java.lang.RuntimeException: java.lang.StringIndexOutOfBoundsException: Some Error\n\tat org.apache.storm.utils.DisruptorQueue.consumeBatchToCursor(DisruptorQueue.java:128)\n\tat org.apache.storm.utils.DisruptorQueue.consumeBatchWhenAvailable(DisruptorQueue.java:99)\n\tat org.apache.storm.disruptor$consume_batch_when_available.invoke(disruptor.clj:80)\n\tat backtype...more.."
}],
"topologyId": "WordCount3-1-1402960825",
"tasks": 5,
"window": "600",
"profilerActive": [
{
"host": "10.11.1.70",
"port": "6701",
"dumplink":"http:\/\/10.11.1.70:8000\/dumps\/ex-1-1452718803\/10.11.1.70%3A6701",
"timestamp":"576328"
}
],
"profilingAndDebuggingCapable": true,
"profileActionEnabled": true,
"spoutSummary": [
{
"windowPretty": "10m 0s",
"window": "600",
"emitted": 28500,
"transferred": 28460,
"completeLatency": "0.000",
"acked": 0,
"failed": 0
},
{
"windowPretty": "3h 0m 0s",
"window": "10800",
"emitted": 127640,
"transferred": 127440,
"completeLatency": "0.000",
"acked": 0,
"failed": 0
},
{
"windowPretty": "1d 0h 0m 0s",
"window": "86400",
"emitted": 127640,
"transferred": 127440,
"completeLatency": "0.000",
"acked": 0,
"failed": 0
},
{
"windowPretty": "All time",
"window": ":all-time",
"emitted": 127640,
"transferred": 127440,
"completeLatency": "0.000",
"acked": 0,
"failed": 0
}
],
"outputStats": [
{
"stream": "__metrics",
"emitted": 40,
"transferred": 0,
"completeLatency": "0",
"acked": 0,
"failed": 0
},
{
"stream": "default",
"emitted": 28460,
"transferred": 28460,
"completeLatency": "0",
"acked": 0,
"failed": 0
}
],
"executorStats": [
{
"workerLogLink": "http://10.11.1.7:8000/log?file=worker-6701.log",
"emitted": 5720,
"port": 6701,
"completeLatency": "0.000",
"transferred": 5720,
"host": "10.11.1.7",
"acked": 0,
"uptime": "43m 4s",
"uptimeSeconds": 2584,
"id": "[24-24]",
"failed": 0
},
{
"workerLogLink": "http://10.11.1.7:8000/log?file=worker-6703.log",
"emitted": 5700,
"port": 6703,
"completeLatency": "0.000",
"transferred": 5700,
"host": "10.11.1.7",
"acked": 0,
"uptime": "42m 57s",
"uptimeSeconds": 2577,
"id": "[25-25]",
"failed": 0
},
{
"workerLogLink": "http://10.11.1.7:8000/log?file=worker-6702.log",
"emitted": 5700,
"port": 6702,
"completeLatency": "0.000",
"transferred": 5680,
"host": "10.11.1.7",
"acked": 0,
"uptime": "42m 57s",
"uptimeSeconds": 2577,
"id": "[26-26]",
"failed": 0
},
{
"workerLogLink": "http://10.11.1.7:8000/log?file=worker-6701.log",
"emitted": 5700,
"port": 6701,
"completeLatency": "0.000",
"transferred": 5680,
"host": "10.11.1.7",
"acked": 0,
"uptime": "43m 4s",
"uptimeSeconds": 2584,
"id": "[27-27]",
"failed": 0
},
{
"workerLogLink": "http://10.11.1.7:8000/log?file=worker-6703.log",
"emitted": 5680,
"port": 6703,
"completeLatency": "0.000",
"transferred": 5680,
"host": "10.11.1.7",
"acked": 0,
"uptime": "42m 57s",
"uptimeSeconds": 2577,
"id": "[28-28]",
"failed": 0
}
]
}
Request to start profiler on worker with timeout. Returns status and link to profiler artifacts for worker.
Parameter | Value | Description |
---|---|---|
id | String (required) | Topology Id |
host-port | String (required) | Worker Id |
timeout | String (required) | Time out for profiler to stop in minutes |
Response fields:
Field | Value | Description |
---|---|---|
id | String | Worker id |
status | String | Response Status |
timeout | String | Requested timeout |
dumplink | String | Link to logviewer URL for worker profiler documents. |
Examples:
1. http://ui-daemon-host-name:8080/api/v1/topology/wordcount-1-1446614150/profiling/start/10.11.1.7:6701/10
2. http://ui-daemon-host-name:8080/api/v1/topology/wordcount-1-1446614150/profiling/start/10.11.1.7:6701/5
3. http://ui-daemon-host-name:8080/api/v1/topology/wordcount-1-1446614150/profiling/start/10.11.1.7:6701/20
Sample response:
{
"status": "ok",
"id": "10.11.1.7:6701",
"timeout": "10",
"dumplink": "http:\/\/10.11.1.7:8000\/dumps\/wordcount-1-1446614150\/10.11.1.7%3A6701"
}
Request to dump profiler recording on worker. Returns status and worker id for the request.
Parameter | Value | Description |
---|---|---|
id | String (required) | Topology Id |
host-port | String (required) | Worker Id |
Response fields:
Field | Value | Description |
---|---|---|
id | String | Worker id |
status | String | Response Status |
Examples:
1. http://ui-daemon-host-name:8080/api/v1/topology/wordcount-1-1446614150/profiling/dumpprofile/10.11.1.7:6701
Sample response:
{
"status": "ok",
"id": "10.11.1.7:6701",
}
Request to stop profiler on worker. Returns status and worker id for the request.
Parameter | Value | Description |
---|---|---|
id | String (required) | Topology Id |
host-port | String (required) | Worker Id |
Response fields:
Field | Value | Description |
---|---|---|
id | String | Worker id |
status | String | Response Status |
Examples:
1. http://ui-daemon-host-name:8080/api/v1/topology/wordcount-1-1446614150/profiling/stop/10.11.1.7:6701
Sample response:
{
"status": "ok",
"id": "10.11.1.7:6701",
}
Request to dump jstack on worker. Returns status and worker id for the request.
Parameter | Value | Description |
---|---|---|
id | String (required) | Topology Id |
host-port | String (required) | Worker Id |
Response fields:
Field | Value | Description |
---|---|---|
id | String | Worker id |
status | String | Response Status |
Examples:
1. http://ui-daemon-host-name:8080/api/v1/topology/wordcount-1-1446614150/profiling/dumpjstack/10.11.1.7:6701
Sample response:
{
"status": "ok",
"id": "10.11.1.7:6701",
}
Request to dump heap (jmap) on worker. Returns status and worker id for the request.
Parameter | Value | Description |
---|---|---|
id | String (required) | Topology Id |
host-port | String (required) | Worker Id |
Response fields:
Field | Value | Description |
---|---|---|
id | String | Worker id |
status | String | Response Status |
Examples:
1. http://ui-daemon-host-name:8080/api/v1/topology/wordcount-1-1446614150/profiling/dumpheap/10.11.1.7:6701
Sample response:
{
"status": "ok",
"id": "10.11.1.7:6701",
}
Request to request the worker. Returns status and worker id for the request.
Parameter | Value | Description |
---|---|---|
id | String (required) | Topology Id |
host-port | String (required) | Worker Id |
Response fields:
Field | Value | Description |
---|---|---|
id | String | Worker id |
status | String | Response Status |
Examples:
1. http://ui-daemon-host-name:8080/api/v1/topology/wordcount-1-1446614150/profiling/restartworker/10.11.1.7:6701
Sample response:
{
"status": "ok",
"id": "10.11.1.7:6701",
}
Activates a topology.
Parameter | Value | Description |
---|---|---|
id | String (required) | Topology Id |
Sample Response:
{"topologyOperation":"activate","topologyId":"wordcount-1-1420308665","status":"success"}
Deactivates a topology.
Parameter | Value | Description |
---|---|---|
id | String (required) | Topology Id |
Sample Response:
{"topologyOperation":"deactivate","topologyId":"wordcount-1-1420308665","status":"success"}
Rebalances a topology.
Parameter | Value | Description |
---|---|---|
id | String (required) | Topology Id |
wait-time | String (required) | Wait time before rebalance happens |
rebalanceOptions | Json (optional) | topology rebalance options |
Sample rebalanceOptions json:
{"rebalanceOptions" : {"numWorkers" : 2, "executors" : {"spout" :4, "count" : 10}}, "callback" : "foo"}
Examples:
curl -i -b ~/cookiejar.txt -c ~/cookiejar.txt -X POST
-H "Content-Type: application/json"
-d '{"rebalanceOptions": {"numWorkers": 2, "executors": { "spout" : "5", "split": 7, "count": 5 }}, "callback":"foo"}'
http://localhost:8080/api/v1/topology/wordcount-1-1420308665/rebalance/0
Sample Response:
{"topologyOperation":"rebalance","topologyId":"wordcount-1-1420308665","status":"success"}
Kills a topology.
Parameter | Value | Description |
---|---|---|
id | String (required) | Topology Id |
wait-time | String (required) | Wait time before rebalance happens |
Caution: Small wait times (0-5 seconds) may increase the probability of triggering the bug reported in STORM-112, which may result in broker Supervisor daemons.
Sample Response:
{"topologyOperation":"kill","topologyId":"wordcount-1-1420308665","status":"success"}
The API returns 500 HTTP status codes in case of any errors.
Sample response:
{
"error": "Internal Server Error",
"errorMessage": "java.lang.NullPointerException\n\tat clojure.core$name.invoke(core.clj:1505)\n\tat org.apache.storm.ui.core$component_page.invoke(core.clj:752)\n\tat org.apache.storm.ui.core$fn__7766.invoke(core.clj:782)\n\tat compojure.core$make_route$fn__5755.invoke(core.clj:93)\n\tat compojure.core$if_route$fn__5743.invoke(core.clj:39)\n\tat compojure.core$if_method$fn__5736.invoke(core.clj:24)\n\tat compojure.core$routing$fn__5761.invoke(core.clj:106)\n\tat clojure.core$some.invoke(core.clj:2443)\n\tat compojure.core$routing.doInvoke(core.clj:106)\n\tat clojure.lang.RestFn.applyTo(RestFn.java:139)\n\tat clojure.core$apply.invoke(core.clj:619)\n\tat compojure.core$routes$fn__5765.invoke(core.clj:111)\n\tat ring.middleware.reload$wrap_reload$fn__6880.invoke(reload.clj:14)\n\tat org.apache.storm.ui.core$catch_errors$fn__7800.invoke(core.clj:836)\n\tat ring.middleware.keyword_params$wrap_keyword_params$fn__6319.invoke(keyword_params.clj:27)\n\tat ring.middleware.nested_params$wrap_nested_params$fn__6358.invoke(nested_params.clj:65)\n\tat ring.middleware.params$wrap_params$fn__6291.invoke(params.clj:55)\n\tat ring.middleware.multipart_params$wrap_multipart_params$fn__6386.invoke(multipart_params.clj:103)\n\tat ring.middleware.flash$wrap_flash$fn__6675.invoke(flash.clj:14)\n\tat ring.middleware.session$wrap_session$fn__6664.invoke(session.clj:43)\n\tat ring.middleware.cookies$wrap_cookies$fn__6595.invoke(cookies.clj:160)\n\tat ring.adapter.jetty$proxy_handler$fn__6112.invoke(jetty.clj:16)\n\tat ring.adapter.jetty.proxy$org.mortbay.jetty.handler.AbstractHandler$0.handle(Unknown Source)\n\tat org.mortbay.jetty.handler.HandlerWrapper.handle(HandlerWrapper.java:152)\n\tat org.mortbay.jetty.Server.handle(Server.java:326)\n\tat org.mortbay.jetty.HttpConnection.handleRequest(HttpConnection.java:542)\n\tat org.mortbay.jetty.HttpConnection$RequestHandler.headerComplete(HttpConnection.java:928)\n\tat org.mortbay.jetty.HttpParser.parseNext(HttpParser.java:549)\n\tat org.mortbay.jetty.HttpParser.parseAvailable(HttpParser.java:212)\n\tat org.mortbay.jetty.HttpConnection.handle(HttpConnection.java:404)\n\tat org.mortbay.jetty.bio.SocketConnector$Connection.run(SocketConnector.java:228)\n\tat org.mortbay.thread.QueuedThreadPool$PoolThread.run(QueuedThreadPool.java:582)\n"
}