Rob Fellows created NIFI-15868:
----------------------------------
Summary: Stop connector API response returns stale
availableActions; DRAIN_FLOWFILES reported not-allowed even when queues have
data
Key: NIFI-15868
URL: https://issues.apache.org/jira/browse/NIFI-15868
Project: Apache NiFi
Issue Type: Bug
Components: Core Framework
Reporter: Rob Fellows
When stopping a connector via {{PUT /nifi-api/connectors/\{id}/run-status}}
with state {{{}STOPPED{}}}, the response {{ConnectorEntity.availableActions}}
reports {{DRAIN_FLOWFILES}} (and {{{}PURGE_FLOWFILES{}}}, {{{}DELETE{}}}) as
{{allowed=false}} with reason {{{}"Connector must be stopped"{}}}, even when
flowfiles are queued. A follow-up {{GET /nifi-api/connectors/\{id}}} returns
the same connector with {{{}DRAIN_FLOWFILES.allowed=true{}}}. UIs that rely on
the stop response to enable/disable actions therefore show DRAIN as disabled
until the user manually refreshes.
Root cause {{StandardConnectorNode.stop(FlowEngine)}} is asynchronous: it sets
{{{}desiredState=STOPPED{}}}, transitions {{currentState}} to {{{}STOPPING{}}},
schedules the actual stop on the lifecycle executor, and returns a
{{Future<Void>}} immediately. {{StandardConnectorDAO.stopConnector(String)}}
discards that future, so {{StandardNiFiServiceFacade.scheduleConnector(...)}}
builds the response DTO while the connector's {{currentState}} is still
{{{}STOPPING{}}}. {{getAvailableActions()}} keys off {{isStopped()}} (which
requires {{{}currentState == STOPPED{}}}), so {{DRAIN_FLOWFILES}} is computed
as not-allowed for the response.
Steps to reproduce
# Start a running connector that has flowfiles queued in its internal process
group.
# From the connector canvas, stop the connector and capture the response of
{{{}PUT /nifi-api/connectors/\{id}/run-status{}}}.
# Observe {{availableActions}} for {{DRAIN_FLOWFILES}} — {{{}allowed:
false{}}}, {{{}reason: "Connector must be stopped"{}}}.
# Reload the page (or call {{{}GET /nifi-api/connectors/\{id}{}}}) —
{{DRAIN_FLOWFILES}} is now {{{}allowed: true{}}}.
Expected The 200 response from the stop call reflects the post-stop state, so
{{availableActions}} matches what a follow-up GET would return.
Affected code
* {{StandardNiFiServiceFacade.scheduleConnector}} (~line 3674)
* {{StandardConnectorDAO.stopConnector}} (~line 133)
* {{StandardConnectorRepository.stopConnector}} (~line 478)
* {{StandardConnectorNode.stop}} (~line 433); state transitions to {{STOPPED}}
only inside {{stopComponent}} (~line 589)
* {{StandardConnectorNode.getAvailableActions}} / {{isStopped}} (~lines 1145,
1164)
{*}Related{*}: start path has the same async pattern but is benign
{{StandardConnectorDAO.startConnector}} is also fire-and-forget and the
response is built while the connector is in {{STARTING}} (not {{{}RUNNING{}}}).
The {{availableActions}} for {{STARTING}} happen to match those for {{RUNNING}}
because {{createStopAction}} explicitly treats {{STARTING}} like {{RUNNING}}
({{{}StandardConnectorNode.java{}}}, line ~1201). The only user-visible
artifact is that {{component.state}} briefly reports {{STARTING}} in the start
response. The stop path does not have an analogous "treat {{STOPPING}} as
{{{}STOPPED{}}}" branch, which is why DRAIN/PURGE/DELETE come back wrongly
disabled. The cleanest fix — having {{scheduleConnector}} await the lifecycle
{{Future}} before serializing the response — would correct both paths
consistently.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)