From 0251840c723223659d054efcbaf41642eca05f54 Mon Sep 17 00:00:00 2001
From: Guido Petretto <g.petretto@gmail.com>
Date: Wed, 21 Feb 2024 20:43:07 +0100
Subject: [PATCH] map all slurm states

---
 src/qtoolkit/core/data_objects.py | 14 ++++++++++++++
 src/qtoolkit/io/slurm.py          | 24 +++++++++++++++++++++++-
 2 files changed, 37 insertions(+), 1 deletion(-)

diff --git a/src/qtoolkit/core/data_objects.py b/src/qtoolkit/core/data_objects.py
index 6a0f440..c06e4fe 100644
--- a/src/qtoolkit/core/data_objects.py
+++ b/src/qtoolkit/core/data_objects.py
@@ -58,6 +58,20 @@ class QState(QTKEnum):
     queue manager (e.g. PBS, SLURM, ...) needs to be
     defined.
 
+    UNDETERMINED: The job status cannot be determined. This is a permanent
+        issue, not being solvable by asking again for the job state.
+    QUEUED: The job is queued for being scheduled and executed.
+    QUEUED HELD: The job has been placed on hold by the system, the
+        administrator, or the submitting user.
+    RUNNING: The job is running on an execution host.
+    SUSPENDED: The job has been suspended by the user, the system or the
+        administrator.
+    REQUEUED: The job was re-queued by the DRM system, and is eligible to run.
+    REQUEUED HELD: The job was re-queued by the DRM system, and is currently
+        placed on hold by the system, the administrator, or the submitting user.
+    DONE: The job finished without an error.
+    FAILED: The job exited abnormally before finishing.
+
     Note that not all these standardized states are available in the
     actual queue manager implementations.
     """
diff --git a/src/qtoolkit/io/slurm.py b/src/qtoolkit/io/slurm.py
index 8254f2a..08f48d3 100644
--- a/src/qtoolkit/io/slurm.py
+++ b/src/qtoolkit/io/slurm.py
@@ -88,6 +88,7 @@
 
 
 class SlurmState(QSubState):
+    BOOT_FAIL = "BOOT_FAIL", "BF"
     CANCELLED = "CANCELLED", "CA"
     COMPLETING = "COMPLETING", "CG"
     COMPLETED = "COMPLETED", "CD"
@@ -97,7 +98,17 @@ class SlurmState(QSubState):
     NODE_FAIL = "NODE_FAIL", "NF"
     OUT_OF_MEMORY = "OUT_OF_MEMORY", "OOM"
     PENDING = "PENDING", "PD"
+    PREEMPTED = "PREEMPTED", "PR"
+    RESV_DEL_HOLD = "RESV_DEL_HOLD", "RD"
+    REQUEUE_FED = "REQUEUE_FED", "RF"
+    REQUEUE_HOLD = "REQUEUE_HOLD", "RH"
+    RESIZING = "RESIZING", "RS"
+    REVOKED = "REVOKED", "RV"
     RUNNING = "RUNNING", "R"
+    SIGNALING = "SIGNALING", "SI"
+    SPECIAL_EXIT = "SPECIAL_EXIT", "SE"
+    STAGE_OUT = "STAGE_OUT", "SO"
+    STOPPED = "STOPPED", "ST"
     SUSPENDED = "SUSPENDED", "S"
     TIMEOUT = "TIMEOUT", "TO"
 
@@ -108,7 +119,8 @@ def qstate(self) -> QState:
 
 
 _STATUS_MAPPING = {
-    SlurmState.CANCELLED: QState.SUSPENDED,  # Should this be failed ?
+    SlurmState.BOOT_FAIL: QState.FAILED,
+    SlurmState.CANCELLED: QState.FAILED,
     SlurmState.COMPLETING: QState.RUNNING,
     SlurmState.COMPLETED: QState.DONE,
     SlurmState.CONFIGURING: QState.QUEUED,
@@ -117,7 +129,17 @@ def qstate(self) -> QState:
     SlurmState.NODE_FAIL: QState.FAILED,
     SlurmState.OUT_OF_MEMORY: QState.FAILED,
     SlurmState.PENDING: QState.QUEUED,
+    SlurmState.PREEMPTED: QState.FAILED,
+    SlurmState.RESV_DEL_HOLD: QState.QUEUED_HELD,
+    SlurmState.REQUEUE_FED: QState.QUEUED,
+    SlurmState.REQUEUE_HOLD: QState.QUEUED,
+    SlurmState.RESIZING: QState.RUNNING,
+    SlurmState.REVOKED: QState.FAILED,
     SlurmState.RUNNING: QState.RUNNING,
+    SlurmState.SIGNALING: QState.RUNNING,
+    SlurmState.SPECIAL_EXIT: QState.FAILED,
+    SlurmState.STAGE_OUT: QState.RUNNING,
+    SlurmState.STOPPED: QState.RUNNING,
     SlurmState.SUSPENDED: QState.SUSPENDED,
     SlurmState.TIMEOUT: QState.FAILED,
 }