Rajmund Takacs created NIFI-13829:
-------------------------------------
Summary: MonitorActivity processor generating false inactivity
files
Key: NIFI-13829
URL: https://issues.apache.org/jira/browse/NIFI-13829
Project: Apache NiFi
Issue Type: Bug
Components: Extensions
Affects Versions: 2.0.0-M4, 1.27.0
Reporter: Rajmund Takacs
Assignee: Rajmund Takacs
MonitorActivity generates false inactivity reports (with big ratio), if the
FlowFiles are coming less frequent, than the 2/3 of the configured threshold.
This happens only in cluster monitoring scope, where only primary node reports,
and only non-primary nodes have FlowFiles to process.
This is because, the processor synchronizes the common state only 3 times
within the threshold: 1T/3, 2T/3, 3T/3. If the FlowFile hits the non-primary
node between the second (2T/3) and the third sync point (3T/3), then the state
from the non-primary node will only be shared with the primary node, when the
last sync point comes. And as sync points are relativeley calculated by each
node, there is a race condition in terms of which happens first: the primary
node reads the common state, or the non-primary node writes the new state. In
case of the former, primary node will generate a false indication.
{noformat}
[PRIMARY] - SY0 ---- SY1 ----- SY2 ---- SY3 ---- SY4 ----
/ / / / ? /
[C STATE] ------X---:---------:--------:-------:X-:------
/ / / /
[SLAVE N] -- FF1 ---- SY1 ---- SY2 -- FF2 -- SY3 --------
{noformat}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
