From 1fa4f029d76556c215b181ba1ae39a35e02402f3 Mon Sep 17 00:00:00 2001
From: Oliver Sanders <oliver.sanders@metoffice.gov.uk>
Date: Mon, 28 Oct 2019 11:44:54 +0000
Subject: [PATCH] retry status: fixes

---
 bin/cylc-submit                   | 13 ++++++-------
 cylc/flow/task_events_mgr.py      | 25 ++++++++++++++++++++++---
 tests/retries/03-upgrade.t        |  9 ++++++---
 tests/retries/03-upgrade/suite.rc |  4 ++--
 4 files changed, 36 insertions(+), 15 deletions(-)

diff --git a/bin/cylc-submit b/bin/cylc-submit
index cc61af73713..0caedf238a8 100755
--- a/bin/cylc-submit
+++ b/bin/cylc-submit
@@ -59,6 +59,7 @@ from cylc.flow.task_proxy import TaskProxy
 from cylc.flow.task_state import TASK_STATUS_SUBMIT_FAILED
 from cylc.flow.templatevars import load_template_vars
 from cylc.flow.terminal import cli_function
+from cylc.flow.xtrigger_mgr import XtriggerManager
 
 
 def get_option_parser():
@@ -117,25 +118,23 @@ def main(parser, options, suite, *task_ids):
     extract_resources(
         get_suite_srv_dir(suite),
         ['etc/job.sh'])
-    pool = SubProcPool()
+    proc_pool = SubProcPool()
     owner = get_user()
     job_pool = JobPool(suite, owner)
     db_mgr = SuiteDatabaseManager()
-    bdcst_mgr = BroadcastMgr(db_mgr),
+    bdcst_mgr = BroadcastMgr(db_mgr)
     task_job_mgr = TaskJobManager(
         suite,
-        pool,
+        proc_pool,
         db_mgr,
         TaskEventsManager(
             suite,
-            pool,
+            proc_pool,
             db_mgr,
             bdcst_mgr,
             XtriggerManager(
                 suite,
-                owner,
-                bdcst_mgr,
-                proc_pool
+                owner
             ),
             job_pool
         ),
diff --git a/cylc/flow/task_events_mgr.py b/cylc/flow/task_events_mgr.py
index a562aef2540..d20f9965285 100644
--- a/cylc/flow/task_events_mgr.py
+++ b/cylc/flow/task_events_mgr.py
@@ -489,9 +489,28 @@ def _process_message_check(
                 itask, itask.state, self.FLAG_RECEIVED_IGNORED, message,
                 event_time, submit_num, itask.submit_num)
             return False
-        if itask.state(TASK_STATUS_WAITING):
-            # Ignore polled messages if waiting
-            # (this includes "retrying" states)
+
+        if (
+                itask.state(TASK_STATUS_WAITING)
+                and
+                (
+                    (
+                        # task has automatically submit-retried at least once
+                        TimerFlags.SUBMISSION_RETRY in itask.try_timers
+                        and itask.try_timers[
+                            TimerFlags.SUBMISSION_RETRY].num > 0
+                    )
+                    or
+                    (
+                        # task has automatically execn-retried at least once
+                        TimerFlags.EXECUTION_RETRY in itask.try_timers
+                        and itask.try_timers[
+                            TimerFlags.EXECUTION_RETRY].num > 0
+                    )
+                )
+
+        ):
+            # Ignore polled messages if task is already in retrying statuses
             LOG.warning(
                 logfmt,
                 itask, itask.state, self.FLAG_POLLED_IGNORED, message,
diff --git a/tests/retries/03-upgrade.t b/tests/retries/03-upgrade.t
index 847487270bf..3e6c75038e0 100644
--- a/tests/retries/03-upgrade.t
+++ b/tests/retries/03-upgrade.t
@@ -19,7 +19,7 @@
 # Test the upgrade of the old *retrying states to the new xtrigger based
 # retry mechanism.
 . "$(dirname "$0")/test_header"
-set_test_number 7
+set_test_number 10
 install_suite
 
 # install the cylc7 restart database
@@ -35,8 +35,11 @@ log_scan "${TEST_NAME_BASE}-retries" "${FILE}" 30 0.5 \
     '\[b.1\] -submit-num=02' \
     '\[b.1\] status=running: (received)failed/EXIT.*job(02)' \
     '\[b.1\] -job(02) failed, retrying in PT2S' \
-    'xtrigger satisfied: cylc_retry_b.1'
+    'xtrigger satisfied: cylc_retry_b.1' \
+    '\[b.1\] -submit-num=03' \
+    '\[b.1] status=running: (received)succeeded' \
+    '\[c.1] status=running: (received)succeeded'
 
-poll_suite_stopped
+purge_suite "${SUITE_NAME}"
 
 exit
diff --git a/tests/retries/03-upgrade/suite.rc b/tests/retries/03-upgrade/suite.rc
index 3da90c2b7d9..33da6a39e2b 100644
--- a/tests/retries/03-upgrade/suite.rc
+++ b/tests/retries/03-upgrade/suite.rc
@@ -14,6 +14,6 @@
 [runtime]
     [[b]]
         # fail four times then pass
-        script = test "$CYLC_TASK_SUBMIT_NUMBER" -ge 5;
+        script = test "$CYLC_TASK_SUBMIT_NUMBER" -ge 3;
         [[[job]]]
-            execution retry delays = 3*PT2S
+            execution retry delays = 2*PT2S