UncountablePythonSDK 0.0.125__py3-none-any.whl → 0.0.127__py3-none-any.whl

uncountable/integration/queue_runner/command_server/protocol/command_server_pb2.py

@@ -18,7 +18,7 @@ from google.protobuf import timestamp_pb2 as google_dot_protobuf_dot_timestamp__
 
 
 DESCRIPTOR = _descriptor_pool.Default().AddSerializedFile(
-    b'\nQuncountable/integration/queue_runner/command_server/protocol/command_server.proto\x1a\x1fgoogle/protobuf/timestamp.proto"E\n\x11\x45nqueueJobRequest\x12\x14\n\x0cjob_ref_name\x18\x01 \x01(\t\x12\x1a\n\x12serialized_payload\x18\x02 \x01(\t"H\n\x10\x45nqueueJobResult\x12\x1b\n\x13successfully_queued\x18\x01 \x01(\x08\x12\x17\n\x0fqueued_job_uuid\x18\x02 \x01(\t"\x14\n\x12\x43heckHealthRequest"$\n\x11\x43heckHealthResult\x12\x0f\n\x07success\x18\x01 \x01(\x08"6\n\x15ListQueuedJobsRequest\x12\x0e\n\x06offset\x18\x01 \x01(\r\x12\r\n\x05limit\x18\x02 \x01(\r"\xe4\x01\n\x14ListQueuedJobsResult\x12\x43\n\x0bqueued_jobs\x18\x01 \x03(\x0b\x32..ListQueuedJobsResult.ListQueuedJobsResultItem\x1a\x86\x01\n\x18ListQueuedJobsResultItem\x12\x0c\n\x04uuid\x18\x01 \x01(\t\x12\x14\n\x0cjob_ref_name\x18\x02 \x01(\t\x12\x14\n\x0cnum_attempts\x18\x03 \x01(\x03\x12\x30\n\x0csubmitted_at\x18\x04 \x01(\x0b\x32\x1a.google.protobuf.Timestamp2\xc3\x01\n\rCommandServer\x12\x35\n\nEnqueueJob\x12\x12.EnqueueJobRequest\x1a\x11.EnqueueJobResult"\x00\x12\x38\n\x0b\x43heckHealth\x12\x13.CheckHealthRequest\x1a\x12.CheckHealthResult"\x00\x12\x41\n\x0eListQueuedJobs\x12\x16.ListQueuedJobsRequest\x1a\x15.ListQueuedJobsResult"\x00\x62\x06proto3'
+    b'\nQuncountable/integration/queue_runner/command_server/protocol/command_server.proto\x1a\x1fgoogle/protobuf/timestamp.proto"E\n\x11\x45nqueueJobRequest\x12\x14\n\x0cjob_ref_name\x18\x01 \x01(\t\x12\x1a\n\x12serialized_payload\x18\x02 \x01(\t"H\n\x10\x45nqueueJobResult\x12\x1b\n\x13successfully_queued\x18\x01 \x01(\x08\x12\x17\n\x0fqueued_job_uuid\x18\x02 \x01(\t"\x1f\n\x0fRetryJobRequest\x12\x0c\n\x04uuid\x18\x01 \x01(\t"F\n\x0eRetryJobResult\x12\x1b\n\x13successfully_queued\x18\x01 \x01(\x08\x12\x17\n\x0fqueued_job_uuid\x18\x02 \x01(\t"\x1a\n\x18VaccuumQueuedJobsRequest"\x19\n\x17VaccuumQueuedJobsResult"\x14\n\x12\x43heckHealthRequest"$\n\x11\x43heckHealthResult\x12\x0f\n\x07success\x18\x01 \x01(\x08"6\n\x15ListQueuedJobsRequest\x12\x0e\n\x06offset\x18\x01 \x01(\r\x12\r\n\x05limit\x18\x02 \x01(\r"\xf4\x01\n\x14ListQueuedJobsResult\x12\x43\n\x0bqueued_jobs\x18\x01 \x03(\x0b\x32..ListQueuedJobsResult.ListQueuedJobsResultItem\x1a\x96\x01\n\x18ListQueuedJobsResultItem\x12\x0c\n\x04uuid\x18\x01 \x01(\t\x12\x14\n\x0cjob_ref_name\x18\x02 \x01(\t\x12\x14\n\x0cnum_attempts\x18\x03 \x01(\x03\x12\x30\n\x0csubmitted_at\x18\x04 \x01(\x0b\x32\x1a.google.protobuf.Timestamp\x12\x0e\n\x06status\x18\x05 \x01(\t2\xc0\x02\n\rCommandServer\x12\x35\n\nEnqueueJob\x12\x12.EnqueueJobRequest\x1a\x11.EnqueueJobResult"\x00\x12/\n\x08RetryJob\x12\x10.RetryJobRequest\x1a\x0f.RetryJobResult"\x00\x12\x38\n\x0b\x43heckHealth\x12\x13.CheckHealthRequest\x1a\x12.CheckHealthResult"\x00\x12\x41\n\x0eListQueuedJobs\x12\x16.ListQueuedJobsRequest\x1a\x15.ListQueuedJobsResult"\x00\x12J\n\x11VaccuumQueuedJobs\x12\x19.VaccuumQueuedJobsRequest\x1a\x18.VaccuumQueuedJobsResult"\x00\x62\x06proto3'
 )
 
 _globals = globals()
@@ -34,16 +34,24 @@ if _descriptor._USE_C_DESCRIPTORS == False:
     _globals["_ENQUEUEJOBREQUEST"]._serialized_end = 187
     _globals["_ENQUEUEJOBRESULT"]._serialized_start = 189
     _globals["_ENQUEUEJOBRESULT"]._serialized_end = 261
-    _globals["_CHECKHEALTHREQUEST"]._serialized_start = 263
-    _globals["_CHECKHEALTHREQUEST"]._serialized_end = 283
-    _globals["_CHECKHEALTHRESULT"]._serialized_start = 285
-    _globals["_CHECKHEALTHRESULT"]._serialized_end = 321
-    _globals["_LISTQUEUEDJOBSREQUEST"]._serialized_start = 323
-    _globals["_LISTQUEUEDJOBSREQUEST"]._serialized_end = 377
-    _globals["_LISTQUEUEDJOBSRESULT"]._serialized_start = 380
-    _globals["_LISTQUEUEDJOBSRESULT"]._serialized_end = 608
-    _globals["_LISTQUEUEDJOBSRESULT_LISTQUEUEDJOBSRESULTITEM"]._serialized_start = 474
-    _globals["_LISTQUEUEDJOBSRESULT_LISTQUEUEDJOBSRESULTITEM"]._serialized_end = 608
-    _globals["_COMMANDSERVER"]._serialized_start = 611
-    _globals["_COMMANDSERVER"]._serialized_end = 806
+    _globals["_RETRYJOBREQUEST"]._serialized_start = 263
+    _globals["_RETRYJOBREQUEST"]._serialized_end = 294
+    _globals["_RETRYJOBRESULT"]._serialized_start = 296
+    _globals["_RETRYJOBRESULT"]._serialized_end = 366
+    _globals["_VACCUUMQUEUEDJOBSREQUEST"]._serialized_start = 368
+    _globals["_VACCUUMQUEUEDJOBSREQUEST"]._serialized_end = 394
+    _globals["_VACCUUMQUEUEDJOBSRESULT"]._serialized_start = 396
+    _globals["_VACCUUMQUEUEDJOBSRESULT"]._serialized_end = 421
+    _globals["_CHECKHEALTHREQUEST"]._serialized_start = 423
+    _globals["_CHECKHEALTHREQUEST"]._serialized_end = 443
+    _globals["_CHECKHEALTHRESULT"]._serialized_start = 445
+    _globals["_CHECKHEALTHRESULT"]._serialized_end = 481
+    _globals["_LISTQUEUEDJOBSREQUEST"]._serialized_start = 483
+    _globals["_LISTQUEUEDJOBSREQUEST"]._serialized_end = 537
+    _globals["_LISTQUEUEDJOBSRESULT"]._serialized_start = 540
+    _globals["_LISTQUEUEDJOBSRESULT"]._serialized_end = 784
+    _globals["_LISTQUEUEDJOBSRESULT_LISTQUEUEDJOBSRESULTITEM"]._serialized_start = 634
+    _globals["_LISTQUEUEDJOBSRESULT_LISTQUEUEDJOBSRESULTITEM"]._serialized_end = 784
+    _globals["_COMMANDSERVER"]._serialized_start = 787
+    _globals["_COMMANDSERVER"]._serialized_end = 1107
 # @@protoc_insertion_point(module_scope)

uncountable/integration/queue_runner/command_server/protocol/command_server_pb2.pyi

@@ -35,6 +35,30 @@ class EnqueueJobResult(_message.Message):
         self, successfully_queued: bool = ..., queued_job_uuid: _Optional[str] = ...
     ) -> None: ...
 
+class RetryJobRequest(_message.Message):
+    __slots__ = ("uuid",)
+    UUID_FIELD_NUMBER: _ClassVar[int]
+    uuid: str
+    def __init__(self, uuid: _Optional[str] = ...) -> None: ...
+
+class RetryJobResult(_message.Message):
+    __slots__ = ("successfully_queued", "queued_job_uuid")
+    SUCCESSFULLY_QUEUED_FIELD_NUMBER: _ClassVar[int]
+    QUEUED_JOB_UUID_FIELD_NUMBER: _ClassVar[int]
+    successfully_queued: bool
+    queued_job_uuid: str
+    def __init__(
+        self, successfully_queued: bool = ..., queued_job_uuid: _Optional[str] = ...
+    ) -> None: ...
+
+class VaccuumQueuedJobsRequest(_message.Message):
+    __slots__ = ()
+    def __init__(self) -> None: ...
+
+class VaccuumQueuedJobsResult(_message.Message):
+    __slots__ = ()
+    def __init__(self) -> None: ...
+
 class CheckHealthRequest(_message.Message):
     __slots__ = ()
     def __init__(self) -> None: ...
@@ -58,21 +82,24 @@ class ListQueuedJobsRequest(_message.Message):
 class ListQueuedJobsResult(_message.Message):
     __slots__ = ("queued_jobs",)
     class ListQueuedJobsResultItem(_message.Message):
-        __slots__ = ("uuid", "job_ref_name", "num_attempts", "submitted_at")
+        __slots__ = ("uuid", "job_ref_name", "num_attempts", "submitted_at", "status")
         UUID_FIELD_NUMBER: _ClassVar[int]
         JOB_REF_NAME_FIELD_NUMBER: _ClassVar[int]
         NUM_ATTEMPTS_FIELD_NUMBER: _ClassVar[int]
         SUBMITTED_AT_FIELD_NUMBER: _ClassVar[int]
+        STATUS_FIELD_NUMBER: _ClassVar[int]
         uuid: str
         job_ref_name: str
         num_attempts: int
         submitted_at: _timestamp_pb2.Timestamp
+        status: str
         def __init__(
             self,
             uuid: _Optional[str] = ...,
             job_ref_name: _Optional[str] = ...,
             num_attempts: _Optional[int] = ...,
             submitted_at: _Optional[_Union[_timestamp_pb2.Timestamp, _Mapping]] = ...,
+            status: _Optional[str] = ...,
         ) -> None: ...
 
     QUEUED_JOBS_FIELD_NUMBER: _ClassVar[int]

uncountable/integration/queue_runner/command_server/protocol/command_server_pb2_grpc.py

@@ -24,6 +24,11 @@ class CommandServerStub(object):
             request_serializer=uncountable_dot_integration_dot_queue__runner_dot_command__server_dot_protocol_dot_command__server__pb2.EnqueueJobRequest.SerializeToString,
             response_deserializer=uncountable_dot_integration_dot_queue__runner_dot_command__server_dot_protocol_dot_command__server__pb2.EnqueueJobResult.FromString,
         )
+        self.RetryJob = channel.unary_unary(
+            "/CommandServer/RetryJob",
+            request_serializer=uncountable_dot_integration_dot_queue__runner_dot_command__server_dot_protocol_dot_command__server__pb2.RetryJobRequest.SerializeToString,
+            response_deserializer=uncountable_dot_integration_dot_queue__runner_dot_command__server_dot_protocol_dot_command__server__pb2.RetryJobResult.FromString,
+        )
         self.CheckHealth = channel.unary_unary(
             "/CommandServer/CheckHealth",
             request_serializer=uncountable_dot_integration_dot_queue__runner_dot_command__server_dot_protocol_dot_command__server__pb2.CheckHealthRequest.SerializeToString,
@@ -34,6 +39,11 @@ class CommandServerStub(object):
             request_serializer=uncountable_dot_integration_dot_queue__runner_dot_command__server_dot_protocol_dot_command__server__pb2.ListQueuedJobsRequest.SerializeToString,
             response_deserializer=uncountable_dot_integration_dot_queue__runner_dot_command__server_dot_protocol_dot_command__server__pb2.ListQueuedJobsResult.FromString,
         )
+        self.VaccuumQueuedJobs = channel.unary_unary(
+            "/CommandServer/VaccuumQueuedJobs",
+            request_serializer=uncountable_dot_integration_dot_queue__runner_dot_command__server_dot_protocol_dot_command__server__pb2.VaccuumQueuedJobsRequest.SerializeToString,
+            response_deserializer=uncountable_dot_integration_dot_queue__runner_dot_command__server_dot_protocol_dot_command__server__pb2.VaccuumQueuedJobsResult.FromString,
+        )
 
 
 class CommandServerServicer(object):
@@ -45,6 +55,12 @@ class CommandServerServicer(object):
         context.set_details("Method not implemented!")
         raise NotImplementedError("Method not implemented!")
 
+    def RetryJob(self, request, context):
+        """Missing associated documentation comment in .proto file."""
+        context.set_code(grpc.StatusCode.UNIMPLEMENTED)
+        context.set_details("Method not implemented!")
+        raise NotImplementedError("Method not implemented!")
+
     def CheckHealth(self, request, context):
         """Missing associated documentation comment in .proto file."""
         context.set_code(grpc.StatusCode.UNIMPLEMENTED)
@@ -57,6 +73,12 @@ class CommandServerServicer(object):
         context.set_details("Method not implemented!")
         raise NotImplementedError("Method not implemented!")
 
+    def VaccuumQueuedJobs(self, request, context):
+        """Missing associated documentation comment in .proto file."""
+        context.set_code(grpc.StatusCode.UNIMPLEMENTED)
+        context.set_details("Method not implemented!")
+        raise NotImplementedError("Method not implemented!")
+
 
 def add_CommandServerServicer_to_server(servicer, server):
     rpc_method_handlers = {
@@ -65,6 +87,11 @@ def add_CommandServerServicer_to_server(servicer, server):
             request_deserializer=uncountable_dot_integration_dot_queue__runner_dot_command__server_dot_protocol_dot_command__server__pb2.EnqueueJobRequest.FromString,
             response_serializer=uncountable_dot_integration_dot_queue__runner_dot_command__server_dot_protocol_dot_command__server__pb2.EnqueueJobResult.SerializeToString,
         ),
+        "RetryJob": grpc.unary_unary_rpc_method_handler(
+            servicer.RetryJob,
+            request_deserializer=uncountable_dot_integration_dot_queue__runner_dot_command__server_dot_protocol_dot_command__server__pb2.RetryJobRequest.FromString,
+            response_serializer=uncountable_dot_integration_dot_queue__runner_dot_command__server_dot_protocol_dot_command__server__pb2.RetryJobResult.SerializeToString,
+        ),
         "CheckHealth": grpc.unary_unary_rpc_method_handler(
             servicer.CheckHealth,
             request_deserializer=uncountable_dot_integration_dot_queue__runner_dot_command__server_dot_protocol_dot_command__server__pb2.CheckHealthRequest.FromString,
@@ -75,6 +102,11 @@ def add_CommandServerServicer_to_server(servicer, server):
             request_deserializer=uncountable_dot_integration_dot_queue__runner_dot_command__server_dot_protocol_dot_command__server__pb2.ListQueuedJobsRequest.FromString,
             response_serializer=uncountable_dot_integration_dot_queue__runner_dot_command__server_dot_protocol_dot_command__server__pb2.ListQueuedJobsResult.SerializeToString,
         ),
+        "VaccuumQueuedJobs": grpc.unary_unary_rpc_method_handler(
+            servicer.VaccuumQueuedJobs,
+            request_deserializer=uncountable_dot_integration_dot_queue__runner_dot_command__server_dot_protocol_dot_command__server__pb2.VaccuumQueuedJobsRequest.FromString,
+            response_serializer=uncountable_dot_integration_dot_queue__runner_dot_command__server_dot_protocol_dot_command__server__pb2.VaccuumQueuedJobsResult.SerializeToString,
+        ),
     }
     generic_handler = grpc.method_handlers_generic_handler(
         "CommandServer", rpc_method_handlers
@@ -115,6 +147,35 @@ class CommandServer(object):
             metadata,
         )
 
+    @staticmethod
+    def RetryJob(
+        request,
+        target,
+        options=(),
+        channel_credentials=None,
+        call_credentials=None,
+        insecure=False,
+        compression=None,
+        wait_for_ready=None,
+        timeout=None,
+        metadata=None,
+    ):
+        return grpc.experimental.unary_unary(
+            request,
+            target,
+            "/CommandServer/RetryJob",
+            uncountable_dot_integration_dot_queue__runner_dot_command__server_dot_protocol_dot_command__server__pb2.RetryJobRequest.SerializeToString,
+            uncountable_dot_integration_dot_queue__runner_dot_command__server_dot_protocol_dot_command__server__pb2.RetryJobResult.FromString,
+            options,
+            channel_credentials,
+            insecure,
+            call_credentials,
+            compression,
+            wait_for_ready,
+            timeout,
+            metadata,
+        )
+
     @staticmethod
     def CheckHealth(
         request,
@@ -172,3 +233,32 @@ class CommandServer(object):
             timeout,
             metadata,
         )
+
+    @staticmethod
+    def VaccuumQueuedJobs(
+        request,
+        target,
+        options=(),
+        channel_credentials=None,
+        call_credentials=None,
+        insecure=False,
+        compression=None,
+        wait_for_ready=None,
+        timeout=None,
+        metadata=None,
+    ):
+        return grpc.experimental.unary_unary(
+            request,
+            target,
+            "/CommandServer/VaccuumQueuedJobs",
+            uncountable_dot_integration_dot_queue__runner_dot_command__server_dot_protocol_dot_command__server__pb2.VaccuumQueuedJobsRequest.SerializeToString,
+            uncountable_dot_integration_dot_queue__runner_dot_command__server_dot_protocol_dot_command__server__pb2.VaccuumQueuedJobsResult.FromString,
+            options,
+            channel_credentials,
+            insecure,
+            call_credentials,
+            compression,
+            wait_for_ready,
+            timeout,
+            metadata,
+        )

uncountable/integration/queue_runner/command_server/types.py

@@ -8,6 +8,8 @@ from uncountable.types import queued_job_t
 
 class CommandType(StrEnum):
     ENQUEUE_JOB = "enqueue_job"
+    RETRY_JOB = "retry_job"
+    VACCUUM_QUEUED_JOBS = "vaccuum_queued_jobs"
 
 
 RT = typing.TypeVar("RT")
@@ -24,6 +26,16 @@ class CommandEnqueueJobResponse:
     queued_job_uuid: str
 
 
+@dataclass(kw_only=True)
+class CommandRetryJobResponse:
+    queued_job_uuid: str | None
+
+
+@dataclass(kw_only=True)
+class CommandVaccuumQueuedJobsResponse:
+    pass
+
+
 @dataclass(kw_only=True)
 class CommandEnqueueJob(CommandBase[CommandEnqueueJobResponse]):
     type: CommandType = CommandType.ENQUEUE_JOB
@@ -32,7 +44,18 @@ class CommandEnqueueJob(CommandBase[CommandEnqueueJobResponse]):
     response_queue: asyncio.Queue[CommandEnqueueJobResponse]
 
 
-_Command = CommandEnqueueJob
+@dataclass(kw_only=True)
+class CommandRetryJob(CommandBase[CommandRetryJobResponse]):
+    type: CommandType = CommandType.RETRY_JOB
+    queued_job_uuid: str
+
+
+@dataclass(kw_only=True)
+class CommandVaccuumQueuedJobs(CommandBase[CommandVaccuumQueuedJobsResponse]):
+    type: CommandType = CommandType.VACCUUM_QUEUED_JOBS
+
+
+_Command = CommandEnqueueJob | CommandRetryJob | CommandVaccuumQueuedJobs
 
 
 CommandQueue = asyncio.Queue[_Command]

uncountable/integration/queue_runner/datastore/datastore_sqlite.py

@@ -2,7 +2,7 @@ import datetime
 import uuid
 from datetime import UTC
 
-from sqlalchemy import delete, insert, select, update
+from sqlalchemy import delete, insert, or_, select, text, update
 from sqlalchemy.engine import Engine
 
 from pkgs.argument_parser import CachedParser
@@ -14,6 +14,8 @@ from uncountable.types import queued_job_t
 
 queued_job_payload_parser = CachedParser(queued_job_t.QueuedJobPayload)
 
+MAX_QUEUE_WINDOW_DAYS = 30
+
 
 class DatastoreSqlite(Datastore):
     def __init__(self, session_maker: DBSessionMaker) -> None:
@@ -23,6 +25,17 @@ class DatastoreSqlite(Datastore):
     @classmethod
     def setup(cls, engine: Engine) -> None:
         Base.metadata.create_all(engine)
+        with engine.connect() as connection:
+            if not bool(
+                connection.execute(
+                    text(
+                        "select exists (select 1 from pragma_table_info('queued_jobs') where name='status');"
+                    )
+                ).scalar()
+            ):
+                connection.execute(
+                    text("alter table queued_jobs add column status VARCHAR")
+                )
 
     def add_job_to_queue(
         self, job_payload: queued_job_t.QueuedJobPayload, job_ref_name: str
@@ -36,6 +49,7 @@ class DatastoreSqlite(Datastore):
                 QueuedJob.id.key: queued_job_uuid,
                 QueuedJob.job_ref_name.key: job_ref_name,
                 QueuedJob.payload.key: serialized_payload,
+                QueuedJob.status.key: queued_job_t.JobStatus.QUEUED,
                 QueuedJob.num_attempts: num_attempts,
                 QueuedJob.submitted_at: submitted_at,
             })
@@ -44,10 +58,48 @@ class DatastoreSqlite(Datastore):
                 queued_job_uuid=queued_job_uuid,
                 job_ref_name=job_ref_name,
                 payload=job_payload,
+                status=queued_job_t.JobStatus.QUEUED,
                 submitted_at=submitted_at,
                 num_attempts=num_attempts,
             )
 
+    def retry_job(
+        self,
+        queued_job_uuid: str,
+    ) -> queued_job_t.QueuedJob | None:
+        with self.session_maker() as session:
+            select_stmt = select(
+                QueuedJob.id,
+                QueuedJob.payload,
+                QueuedJob.num_attempts,
+                QueuedJob.job_ref_name,
+                QueuedJob.status,
+                QueuedJob.submitted_at,
+            ).filter(QueuedJob.id == queued_job_uuid)
+            existing_job = session.execute(select_stmt).one_or_none()
+
+            if (
+                existing_job is None
+                or existing_job.status != queued_job_t.JobStatus.FAILED
+            ):
+                return None
+
+            update_stmt = (
+                update(QueuedJob)
+                .values({QueuedJob.status.key: queued_job_t.JobStatus.QUEUED})
+                .filter(QueuedJob.id == queued_job_uuid)
+            )
+            session.execute(update_stmt)
+
+            return queued_job_t.QueuedJob(
+                queued_job_uuid=existing_job.id,
+                job_ref_name=existing_job.job_ref_name,
+                num_attempts=existing_job.num_attempts,
+                status=queued_job_t.JobStatus.QUEUED,
+                submitted_at=existing_job.submitted_at,
+                payload=queued_job_payload_parser.parse_storage(existing_job.payload),
+            )
+
     def increment_num_attempts(self, queued_job_uuid: str) -> int:
         with self.session_maker() as session:
             update_stmt = (
@@ -68,6 +120,17 @@ class DatastoreSqlite(Datastore):
             delete_stmt = delete(QueuedJob).filter(QueuedJob.id == queued_job_uuid)
             session.execute(delete_stmt)
 
+    def update_job_status(
+        self, queued_job_uuid: str, status: queued_job_t.JobStatus
+    ) -> None:
+        with self.session_maker() as session:
+            update_stmt = (
+                update(QueuedJob)
+                .values({QueuedJob.status.key: status})
+                .filter(QueuedJob.id == queued_job_uuid)
+            )
+            session.execute(update_stmt)
+
     def list_queued_job_metadata(
         self, offset: int = 0, limit: int | None = 100
     ) -> list[queued_job_t.QueuedJobMetadata]:
@@ -77,6 +140,7 @@ class DatastoreSqlite(Datastore):
                     QueuedJob.id,
                     QueuedJob.job_ref_name,
                     QueuedJob.num_attempts,
+                    QueuedJob.status,
                     QueuedJob.submitted_at,
                 )
                 .order_by(QueuedJob.submitted_at)
@@ -89,6 +153,7 @@ class DatastoreSqlite(Datastore):
                     queued_job_uuid=row.id,
                     job_ref_name=row.job_ref_name,
                     num_attempts=row.num_attempts,
+                    status=row.status or queued_job_t.JobStatus.QUEUED,
                     submitted_at=row.submitted_at,
                 )
                 for row in session.execute(select_statement)
@@ -106,9 +171,16 @@ class DatastoreSqlite(Datastore):
                     QueuedJob.payload,
                     QueuedJob.num_attempts,
                     QueuedJob.job_ref_name,
+                    QueuedJob.status,
                     QueuedJob.submitted_at,
                 )
                 .filter(QueuedJob.job_ref_name == job_ref_name)
+                .filter(
+                    or_(
+                        QueuedJob.status == queued_job_t.JobStatus.QUEUED,
+                        QueuedJob.status.is_(None),
+                    )
+                )
                 .limit(1)
                 .order_by(QueuedJob.submitted_at)
             )
@@ -119,6 +191,7 @@ class DatastoreSqlite(Datastore):
                     queued_job_uuid=row.id,
                     job_ref_name=row.job_ref_name,
                     num_attempts=row.num_attempts,
+                    status=row.status or queued_job_t.JobStatus.QUEUED,
                     submitted_at=row.submitted_at,
                     payload=parsed_payload,
                 )
@@ -127,13 +200,23 @@ class DatastoreSqlite(Datastore):
 
     def load_job_queue(self) -> list[queued_job_t.QueuedJob]:
         with self.session_maker() as session:
-            select_stmt = select(
-                QueuedJob.id,
-                QueuedJob.payload,
-                QueuedJob.num_attempts,
-                QueuedJob.job_ref_name,
-                QueuedJob.submitted_at,
-            ).order_by(QueuedJob.submitted_at)
+            select_stmt = (
+                select(
+                    QueuedJob.id,
+                    QueuedJob.payload,
+                    QueuedJob.num_attempts,
+                    QueuedJob.job_ref_name,
+                    QueuedJob.status,
+                    QueuedJob.submitted_at,
+                )
+                .filter(
+                    or_(
+                        QueuedJob.status == queued_job_t.JobStatus.QUEUED,
+                        QueuedJob.status.is_(None),
+                    )
+                )
+                .order_by(QueuedJob.submitted_at)
+            )
 
             queued_jobs: list[queued_job_t.QueuedJob] = []
             for row in session.execute(select_stmt):
@@ -143,9 +226,25 @@ class DatastoreSqlite(Datastore):
                         queued_job_uuid=row.id,
                         job_ref_name=row.job_ref_name,
                         num_attempts=row.num_attempts,
+                        status=row.status or queued_job_t.JobStatus.QUEUED,
                         submitted_at=row.submitted_at,
                         payload=parsed_payload,
                     )
                 )
 
             return queued_jobs
+
+    def vaccuum_queued_jobs(self) -> None:
+        with self.session_maker() as session:
+            delete_stmt = (
+                delete(QueuedJob)
+                .filter(QueuedJob.status == queued_job_t.JobStatus.QUEUED)
+                .filter(
+                    QueuedJob.submitted_at
+                    <= (
+                        datetime.datetime.now(UTC)
+                        - datetime.timedelta(days=MAX_QUEUE_WINDOW_DAYS)
+                    )
+                )
+            )
+            session.execute(delete_stmt)

uncountable/integration/queue_runner/datastore/model.py

@@ -1,7 +1,9 @@
-from sqlalchemy import JSON, BigInteger, Column, DateTime, Text
+from sqlalchemy import JSON, BigInteger, Column, DateTime, Enum, Text
 from sqlalchemy.orm import declarative_base
 from sqlalchemy.sql import func
 
+from uncountable.types import queued_job_t
+
 Base = declarative_base()
 
 
@@ -15,3 +17,8 @@ class QueuedJob(Base):
     )
     payload = Column(JSON, nullable=False)
     num_attempts = Column(BigInteger, nullable=False, default=0, server_default="0")
+    status = Column(
+        Enum(queued_job_t.JobStatus, length=None),
+        default=queued_job_t.JobStatus.QUEUED,
+        nullable=True,
+    )

uncountable/integration/queue_runner/job_scheduler.py

@@ -10,8 +10,13 @@ from uncountable.integration.queue_runner.command_server import (
     CommandEnqueueJob,
     CommandEnqueueJobResponse,
     CommandQueue,
+    CommandRetryJob,
+    CommandRetryJobResponse,
     CommandTask,
 )
+from uncountable.integration.queue_runner.command_server.types import (
+    CommandVaccuumQueuedJobs,
+)
 from uncountable.integration.queue_runner.datastore import DatastoreSqlite
 from uncountable.integration.queue_runner.datastore.interface import Datastore
 from uncountable.integration.queue_runner.worker import Worker
@@ -99,7 +104,9 @@ async def start_scheduler(
                 worker = job_worker_lookup[queued_job.job_ref_name]
             except KeyError as e:
                 logger.log_exception(e)
-                datastore.remove_job_from_queue(queued_job.queued_job_uuid)
+                datastore.update_job_status(
+                    queued_job.queued_job_uuid, queued_job_t.JobStatus.FAILED
+                )
                 return
             await worker.listen_queue.put(queued_job)
 
@@ -135,6 +142,25 @@ async def start_scheduler(
                 CommandEnqueueJobResponse(queued_job_uuid=queued_job_uuid)
             )
 
+        async def _handle_retry_job_command(command: CommandRetryJob) -> None:
+            queued_job = datastore.retry_job(command.queued_job_uuid)
+            if queued_job is None:
+                await command.response_queue.put(
+                    CommandRetryJobResponse(queued_job_uuid=None)
+                )
+                return
+
+            await enqueue_queued_job(queued_job)
+            await command.response_queue.put(
+                CommandRetryJobResponse(queued_job_uuid=queued_job.queued_job_uuid)
+            )
+
+        def _handle_vaccuum_queued_jobs_command(
+            command: CommandVaccuumQueuedJobs,
+        ) -> None:
+            logger.log_info("Vaccuuming queued jobs...")
+            datastore.vaccuum_queued_jobs()
+
         for queued_job in queued_jobs:
             await enqueue_queued_job(queued_job)
 
@@ -151,10 +177,24 @@ async def start_scheduler(
                     match command:
                         case CommandEnqueueJob():
                             await _handle_enqueue_job_command(command=command)
+                        case CommandRetryJob():
+                            await _handle_retry_job_command(command=command)
+                        case CommandVaccuumQueuedJobs():
+                            _handle_vaccuum_queued_jobs_command(command=command)
                         case _:
                             typing.assert_never(command)
                     command_task = asyncio.create_task(command_queue.get())
                 elif task == result_task:
                     queued_job_result = result_task.result()
-                    datastore.remove_job_from_queue(queued_job_result.queued_job_uuid)
+                    match queued_job_result.job_result.success:
+                        case True:
+                            datastore.update_job_status(
+                                queued_job_result.queued_job_uuid,
+                                queued_job_t.JobStatus.SUCCESS,
+                            )
+                        case False:
+                            datastore.update_job_status(
+                                queued_job_result.queued_job_uuid,
+                                queued_job_t.JobStatus.FAILED,
+                            )
                     result_task = asyncio.create_task(result_queue.get())

uncountable/integration/queue_runner/worker.py

@@ -103,13 +103,13 @@ def run_queued_job(
                 profile_metadata=job_details.profile_metadata,
                 logger=job_logger,
                 payload=payload,
+                job_uuid=queued_job.queued_job_uuid,
             )
 
             return execute_job(
                 args=args,
                 profile_metadata=job_details.profile_metadata,
                 job_definition=job_details.job_definition,
-                job_uuid=queued_job.queued_job_uuid,
             )
         except BaseException as e:
             job_logger.log_exception(e)