Revert "Remove slaved id tracker (#14376)" (#14463)

This reverts commit 36097e88c4da51fce6556a58c49bd675f4cf20ab.

7 files changed, 100 insertions, 73 deletions

diff --git a/synapse/storage/databases/main/account_data.py b/synapse/storage/databases/main/account_data.py
index 282687ebce..c38b8a9e5a 100644
--- a/synapse/storage/databases/main/account_data.py
+++ b/synapse/storage/databases/main/account_data.py
@@ -27,6 +27,7 @@ from typing import (
 )
 
 from synapse.api.constants import AccountDataTypes
+from synapse.replication.slave.storage._slaved_id_tracker import SlavedIdTracker
 from synapse.replication.tcp.streams import AccountDataStream, TagAccountDataStream
 from synapse.storage._base import db_to_json
 from synapse.storage.database import (
@@ -67,11 +68,12 @@ class AccountDataWorkerStore(PushRulesWorkerStore, CacheInvalidationWorkerStore)
         # to write account data. A value of `True` implies that `_account_data_id_gen`
         # is an `AbstractStreamIdGenerator` and not just a tracker.
         self._account_data_id_gen: AbstractStreamIdTracker
-        self._can_write_to_account_data = (
-            self._instance_name in hs.config.worker.writers.account_data
-        )
 
         if isinstance(database.engine, PostgresEngine):
+            self._can_write_to_account_data = (
+                self._instance_name in hs.config.worker.writers.account_data
+            )
+
             self._account_data_id_gen = MultiWriterIdGenerator(
                 db_conn=db_conn,
                 db=database,
@@ -93,13 +95,21 @@ class AccountDataWorkerStore(PushRulesWorkerStore, CacheInvalidationWorkerStore)
             # `StreamIdGenerator`, otherwise we use `SlavedIdTracker` which gets
             # updated over replication. (Multiple writers are not supported for
             # SQLite).
-            self._account_data_id_gen = StreamIdGenerator(
-                db_conn,
-                "room_account_data",
-                "stream_id",
-                extra_tables=[("room_tags_revisions", "stream_id")],
-                is_writer=self._instance_name in hs.config.worker.writers.account_data,
-            )
+            if self._instance_name in hs.config.worker.writers.account_data:
+                self._can_write_to_account_data = True
+                self._account_data_id_gen = StreamIdGenerator(
+                    db_conn,
+                    "room_account_data",
+                    "stream_id",
+                    extra_tables=[("room_tags_revisions", "stream_id")],
+                )
+            else:
+                self._account_data_id_gen = SlavedIdTracker(
+                    db_conn,
+                    "room_account_data",
+                    "stream_id",
+                    extra_tables=[("room_tags_revisions", "stream_id")],
+                )
 
         account_max = self.get_max_account_data_stream_id()
         self._account_data_stream_cache = StreamChangeCache(
diff --git a/synapse/storage/databases/main/devices.py b/synapse/storage/databases/main/devices.py
index 3e5c16b15b..aa58c2adc3 100644
--- a/synapse/storage/databases/main/devices.py
+++ b/synapse/storage/databases/main/devices.py
@@ -38,6 +38,7 @@ from synapse.logging.opentracing import (
     whitelisted_homeserver,
 )
 from synapse.metrics.background_process_metrics import wrap_as_background_process
+from synapse.replication.slave.storage._slaved_id_tracker import SlavedIdTracker
 from synapse.replication.tcp.streams._base import DeviceListsStream, UserSignatureStream
 from synapse.storage._base import SQLBaseStore, db_to_json, make_in_list_sql_clause
 from synapse.storage.database import (
@@ -85,19 +86,28 @@ class DeviceWorkerStore(RoomMemberWorkerStore, EndToEndKeyWorkerStore):
     ):
         super().__init__(database, db_conn, hs)
 
-        # In the worker store this is an ID tracker which we overwrite in the non-worker
-        # class below that is used on the main process.
-        self._device_list_id_gen: AbstractStreamIdTracker = StreamIdGenerator(
-            db_conn,
-            "device_lists_stream",
-            "stream_id",
-            extra_tables=[
-                ("user_signature_stream", "stream_id"),
-                ("device_lists_outbound_pokes", "stream_id"),
-                ("device_lists_changes_in_room", "stream_id"),
-            ],
-            is_writer=hs.config.worker.worker_app is None,
-        )
+        if hs.config.worker.worker_app is None:
+            self._device_list_id_gen: AbstractStreamIdTracker = StreamIdGenerator(
+                db_conn,
+                "device_lists_stream",
+                "stream_id",
+                extra_tables=[
+                    ("user_signature_stream", "stream_id"),
+                    ("device_lists_outbound_pokes", "stream_id"),
+                    ("device_lists_changes_in_room", "stream_id"),
+                ],
+            )
+        else:
+            self._device_list_id_gen = SlavedIdTracker(
+                db_conn,
+                "device_lists_stream",
+                "stream_id",
+                extra_tables=[
+                    ("user_signature_stream", "stream_id"),
+                    ("device_lists_outbound_pokes", "stream_id"),
+                    ("device_lists_changes_in_room", "stream_id"),
+                ],
+            )
 
         # Type-ignore: _device_list_id_gen is mixed in from either DataStore (as a
         # StreamIdGenerator) or SlavedDataStore (as a SlavedIdTracker).
diff --git a/synapse/storage/databases/main/events_worker.py b/synapse/storage/databases/main/events_worker.py
index 296e50d661..467d20253d 100644
--- a/synapse/storage/databases/main/events_worker.py
+++ b/synapse/storage/databases/main/events_worker.py
@@ -59,6 +59,7 @@ from synapse.metrics.background_process_metrics import (
     run_as_background_process,
     wrap_as_background_process,
 )
+from synapse.replication.slave.storage._slaved_id_tracker import SlavedIdTracker
 from synapse.replication.tcp.streams import BackfillStream
 from synapse.replication.tcp.streams.events import EventsStream
 from synapse.storage._base import SQLBaseStore, db_to_json, make_in_list_sql_clause
@@ -212,20 +213,26 @@ class EventsWorkerStore(SQLBaseStore):
             # `StreamIdGenerator`, otherwise we use `SlavedIdTracker` which gets
             # updated over replication. (Multiple writers are not supported for
             # SQLite).
-            self._stream_id_gen = StreamIdGenerator(
-                db_conn,
-                "events",
-                "stream_ordering",
-                is_writer=hs.get_instance_name() in hs.config.worker.writers.events,
-            )
-            self._backfill_id_gen = StreamIdGenerator(
-                db_conn,
-                "events",
-                "stream_ordering",
-                step=-1,
-                extra_tables=[("ex_outlier_stream", "event_stream_ordering")],
-                is_writer=hs.get_instance_name() in hs.config.worker.writers.events,
-            )
+            if hs.get_instance_name() in hs.config.worker.writers.events:
+                self._stream_id_gen = StreamIdGenerator(
+                    db_conn,
+                    "events",
+                    "stream_ordering",
+                )
+                self._backfill_id_gen = StreamIdGenerator(
+                    db_conn,
+                    "events",
+                    "stream_ordering",
+                    step=-1,
+                    extra_tables=[("ex_outlier_stream", "event_stream_ordering")],
+                )
+            else:
+                self._stream_id_gen = SlavedIdTracker(
+                    db_conn, "events", "stream_ordering"
+                )
+                self._backfill_id_gen = SlavedIdTracker(
+                    db_conn, "events", "stream_ordering", step=-1
+                )
 
         events_max = self._stream_id_gen.get_current_token()
         curr_state_delta_prefill, min_curr_state_delta_id = self.db_pool.get_cache_dict(
diff --git a/synapse/storage/databases/main/push_rule.py b/synapse/storage/databases/main/push_rule.py
index 12ad44dbb3..8ae10f6127 100644
--- a/synapse/storage/databases/main/push_rule.py
+++ b/synapse/storage/databases/main/push_rule.py
@@ -30,6 +30,7 @@ from typing import (
 
 from synapse.api.errors import StoreError
 from synapse.config.homeserver import ExperimentalConfig
+from synapse.replication.slave.storage._slaved_id_tracker import SlavedIdTracker
 from synapse.replication.tcp.streams import PushRulesStream
 from synapse.storage._base import SQLBaseStore
 from synapse.storage.database import (
@@ -110,14 +111,14 @@ class PushRulesWorkerStore(
     ):
         super().__init__(database, db_conn, hs)
 
-        # In the worker store this is an ID tracker which we overwrite in the non-worker
-        # class below that is used on the main process.
-        self._push_rules_stream_id_gen: AbstractStreamIdTracker = StreamIdGenerator(
-            db_conn,
-            "push_rules_stream",
-            "stream_id",
-            is_writer=hs.config.worker.worker_app is None,
-        )
+        if hs.config.worker.worker_app is None:
+            self._push_rules_stream_id_gen: AbstractStreamIdTracker = StreamIdGenerator(
+                db_conn, "push_rules_stream", "stream_id"
+            )
+        else:
+            self._push_rules_stream_id_gen = SlavedIdTracker(
+                db_conn, "push_rules_stream", "stream_id"
+            )
 
         push_rules_prefill, push_rules_id = self.db_pool.get_cache_dict(
             db_conn,
diff --git a/synapse/storage/databases/main/pusher.py b/synapse/storage/databases/main/pusher.py
index fee37b9ce4..4a01562d45 100644
--- a/synapse/storage/databases/main/pusher.py
+++ b/synapse/storage/databases/main/pusher.py
@@ -27,6 +27,7 @@ from typing import (
 )
 
 from synapse.push import PusherConfig, ThrottleParams
+from synapse.replication.slave.storage._slaved_id_tracker import SlavedIdTracker
 from synapse.replication.tcp.streams import PushersStream
 from synapse.storage._base import SQLBaseStore, db_to_json
 from synapse.storage.database import (
@@ -58,15 +59,20 @@ class PusherWorkerStore(SQLBaseStore):
     ):
         super().__init__(database, db_conn, hs)
 
-        # In the worker store this is an ID tracker which we overwrite in the non-worker
-        # class below that is used on the main process.
-        self._pushers_id_gen: AbstractStreamIdTracker = StreamIdGenerator(
-            db_conn,
-            "pushers",
-            "id",
-            extra_tables=[("deleted_pushers", "stream_id")],
-            is_writer=hs.config.worker.worker_app is None,
-        )
+        if hs.config.worker.worker_app is None:
+            self._pushers_id_gen: AbstractStreamIdTracker = StreamIdGenerator(
+                db_conn,
+                "pushers",
+                "id",
+                extra_tables=[("deleted_pushers", "stream_id")],
+            )
+        else:
+            self._pushers_id_gen = SlavedIdTracker(
+                db_conn,
+                "pushers",
+                "id",
+                extra_tables=[("deleted_pushers", "stream_id")],
+            )
 
         self.db_pool.updates.register_background_update_handler(
             "remove_deactivated_pushers",
diff --git a/synapse/storage/databases/main/receipts.py b/synapse/storage/databases/main/receipts.py
index 64519587f8..dc6989527e 100644
--- a/synapse/storage/databases/main/receipts.py
+++ b/synapse/storage/databases/main/receipts.py
@@ -27,6 +27,7 @@ from typing import (
 )
 
 from synapse.api.constants import EduTypes
+from synapse.replication.slave.storage._slaved_id_tracker import SlavedIdTracker
 from synapse.replication.tcp.streams import ReceiptsStream
 from synapse.storage._base import SQLBaseStore, db_to_json, make_in_list_sql_clause
 from synapse.storage.database import (
@@ -60,9 +61,6 @@ class ReceiptsWorkerStore(SQLBaseStore):
         hs: "HomeServer",
     ):
         self._instance_name = hs.get_instance_name()
-
-        # In the worker store this is an ID tracker which we overwrite in the non-worker
-        # class below that is used on the main process.
         self._receipts_id_gen: AbstractStreamIdTracker
 
         if isinstance(database.engine, PostgresEngine):
@@ -89,12 +87,14 @@ class ReceiptsWorkerStore(SQLBaseStore):
             # `StreamIdGenerator`, otherwise we use `SlavedIdTracker` which gets
             # updated over replication. (Multiple writers are not supported for
             # SQLite).
-            self._receipts_id_gen = StreamIdGenerator(
-                db_conn,
-                "receipts_linearized",
-                "stream_id",
-                is_writer=hs.get_instance_name() in hs.config.worker.writers.receipts,
-            )
+            if hs.get_instance_name() in hs.config.worker.writers.receipts:
+                self._receipts_id_gen = StreamIdGenerator(
+                    db_conn, "receipts_linearized", "stream_id"
+                )
+            else:
+                self._receipts_id_gen = SlavedIdTracker(
+                    db_conn, "receipts_linearized", "stream_id"
+                )
 
         super().__init__(database, db_conn, hs)
 
diff --git a/synapse/storage/util/id_generators.py b/synapse/storage/util/id_generators.py
index 1af0af1266..2dfe4c0b66 100644
--- a/synapse/storage/util/id_generators.py
+++ b/synapse/storage/util/id_generators.py
@@ -186,13 +186,11 @@ class StreamIdGenerator(AbstractStreamIdGenerator):
         column: str,
         extra_tables: Iterable[Tuple[str, str]] = (),
         step: int = 1,
-        is_writer: bool = True,
     ) -> None:
         assert step != 0
         self._lock = threading.Lock()
         self._step: int = step
         self._current: int = _load_current_id(db_conn, table, column, step)
-        self._is_writer = is_writer
         for table, column in extra_tables:
             self._current = (max if step > 0 else min)(
                 self._current, _load_current_id(db_conn, table, column, step)
@@ -206,11 +204,9 @@ class StreamIdGenerator(AbstractStreamIdGenerator):
         self._unfinished_ids: OrderedDict[int, int] = OrderedDict()
 
     def advance(self, instance_name: str, new_id: int) -> None:
-        # Advance should never be called on a writer instance, only over replication
-        if self._is_writer:
-            raise Exception("Replication is not supported by writer StreamIdGenerator")
-
-        self._current = (max if self._step > 0 else min)(self._current, new_id)
+        # `StreamIdGenerator` should only be used when there is a single writer,
+        # so replication should never happen.
+        raise Exception("Replication is not supported by StreamIdGenerator")
 
     def get_next(self) -> AsyncContextManager[int]:
         with self._lock:
@@ -253,9 +249,6 @@ class StreamIdGenerator(AbstractStreamIdGenerator):
         return _AsyncCtxManagerWrapper(manager())
 
     def get_current_token(self) -> int:
-        if self._is_writer:
-            return self._current
-
         with self._lock:
             if self._unfinished_ids:
                 return next(iter(self._unfinished_ids)) - self._step