Remove Postgres 9.4 support

.buildkite/pipeline.yml

@@ -114,23 +114,6 @@ steps:
         - exit_status: 2
           limit: 2
 
-  - label: ":python: 3.5 / :postgres: 9.4"
-    env:
-      TRIAL_FLAGS: "-j 4"
-    command:
-      - "bash -c 'python -m pip install tox && python -m tox -e py35-postgres,codecov'"
-    plugins:
-      - docker-compose#v2.1.0:
-          run: testenv
-          config:
-            - .buildkite/docker-compose.py35.pg94.yaml
-    retry:
-      automatic:
-        - exit_status: -1
-          limit: 2
-        - exit_status: 2
-          limit: 2
-
   - label: ":python: 3.5 / :postgres: 9.5"
     env:
       TRIAL_FLAGS: "-j 4"

CONTRIBUTING.rst

@@ -30,21 +30,20 @@ use github's pull request workflow to review the contribution, and either ask
 you to make any refinements needed or merge it and make them ourselves. The
 changes will then land on master when we next do a release.
 
-We use `CircleCI <https://circleci.com/gh/matrix-org>`_ and `Travis CI
-<https://travis-ci.org/matrix-org/synapse>`_ for continuous integration. All
-pull requests to synapse get automatically tested by Travis and CircleCI.
-If your change breaks the build, this will be shown in GitHub, so please
-keep an eye on the pull request for feedback.
+We use `CircleCI <https://circleci.com/gh/matrix-org>`_ and `Buildkite
+<https://buildkite.com/matrix-dot-org/synapse>`_ for continuous integration.
+Buildkite builds need to be authorised by a maintainer. If your change breaks
+the build, this will be shown in GitHub, so please keep an eye on the pull
+request for feedback.
 
 To run unit tests in a local development environment, you can use:
 
-- ``tox -e py27`` (requires tox to be installed by ``pip install tox``) for
-  SQLite-backed Synapse on Python 2.7.
-- ``tox -e py35`` for SQLite-backed Synapse on Python 3.5.
+- ``tox -e py35`` (requires tox to be installed by ``pip install tox``)
+  for SQLite-backed Synapse on Python 3.5.
 - ``tox -e py36`` for SQLite-backed Synapse on Python 3.6.
-- ``tox -e py27-postgres`` for PostgreSQL-backed Synapse on Python 2.7
+- ``tox -e py36-postgres`` for PostgreSQL-backed Synapse on Python 3.6
   (requires a running local PostgreSQL with access to create databases).
-- ``./test_postgresql.sh`` for PostgreSQL-backed Synapse on Python 2.7
+- ``./test_postgresql.sh`` for PostgreSQL-backed Synapse on Python 3.5
   (requires Docker). Entirely self-contained, recommended if you don't want to
   set up PostgreSQL yourself.
 

UPGRADE.rst

@@ -49,6 +49,33 @@ returned by the Client-Server API:
     # configured on port 443.
     curl -kv https://<host.name>/_matrix/client/versions 2>&1 | grep "Server:"
 
+Upgrading to v1.1
+=================
+
+Synapse 1.1 removes support for older Python and PostgreSQL versions, as
+outlined in `our deprecation notice <https://matrix.org/blog/2019/04/08/synapse-deprecating-postgres-9-4-and-python-2-x>`_.
+
+Minimum Python Version
+----------------------
+
+Synapse v1.1 has a minimum Python requirement of Python 3.5. Python 3.6 or
+Python 3.7 are recommended as they have improved internal string handling,
+significantly reducing memory usage.
+
+If you use current versions of the Matrix.org-distributed Debian packages or
+Docker images, action is not required.
+
+If you install Synapse in a Python virtual environment, please see "Upgrading to
+v0.34.0" for notes on setting up a new virtualenv under Python 3.
+
+Minimum PostgreSQL Version
+--------------------------
+
+If using PostgreSQL under Synapse, you will need to use PostgreSQL 9.5 or above.
+Please see the
+`PostgreSQL documentation <https://www.postgresql.org/docs/11/upgrading.html>`_
+for more details on upgrading your database.
+
 Upgrading to v1.0
 =================
 
@@ -71,11 +98,11 @@ server in a closed federation. This can be done in one of two ways:-
 * Configure a whitelist of server domains to trust via ``federation_certificate_verification_whitelist``.
 
 See the `sample configuration file <docs/sample_config.yaml>`_
-for more details on these settings. 
+for more details on these settings.
 
 Email
 -----
-When a user requests a password reset, Synapse will send an email to the 
+When a user requests a password reset, Synapse will send an email to the
 user to confirm the request.
 
 Previous versions of Synapse delegated the job of sending this email to an

changelog.d/5448.removal

@@ -0,0 +1 @@
+PostgreSQL 9.4 is no longer supported. Synapse requires Postgres 9.5+ or above for Postgres support.

docker/Dockerfile-pgtests

@@ -3,10 +3,10 @@
 FROM matrixdotorg/sytest:latest
 
 # The Sytest image doesn't come with python, so install that
-RUN apt-get -qq install -y python python-dev python-pip
+RUN apt-get update && apt-get -qq install -y python3 python3-dev python3-pip
 
 # We need tox to run the tests in run_pg_tests.sh
-RUN pip install tox
+RUN python3 -m pip install tox
 
 ADD run_pg_tests.sh /pg_tests.sh
 ENTRYPOINT /pg_tests.sh

docker/run_pg_tests.sh

@@ -17,4 +17,4 @@ su -c '/usr/lib/postgresql/9.6/bin/pg_ctl -w -D /var/lib/postgresql/data start'
 # Run the tests
 cd /src
 export TRIAL_FLAGS="-j 4"
-tox --workdir=/tmp -e py27-postgres
+tox --workdir=/tmp -e py35-postgres

docs/postgres.rst

@@ -1,7 +1,7 @@
 Using Postgres
 --------------
 
-Postgres version 9.4 or later is known to work.
+Postgres version 9.5 or later is known to work.
 
 Install postgres client libraries
 =================================
@@ -16,7 +16,7 @@ a postgres database.
 * For other pre-built packages, please consult the documentation from the
   relevant package.
 
-* If you installed synapse `in a virtualenv 
+* If you installed synapse `in a virtualenv
   <../INSTALL.md#installing-from-source>`_, you can install the library with::
 
       ~/synapse/env/bin/pip install matrix-synapse[postgres]

synapse/storage/engines/postgres.py

@@ -45,6 +45,10 @@ def on_new_connection(self, db_conn):
         # together. For example, version 8.1.5 will be returned as 80105
         self._version = db_conn.server_version
 
+        # Are we on a supported PostgreSQL version?
+        if self._version < 90500:
+            raise RuntimeError("Synapse requires PostgreSQL 9.5+ or above.")
+
         db_conn.set_isolation_level(
             self.module.extensions.ISOLATION_LEVEL_REPEATABLE_READ
         )
@@ -64,9 +68,9 @@ def on_new_connection(self, db_conn):
     @property
     def can_native_upsert(self):
         """
-        Can we use native UPSERTs? This requires PostgreSQL 9.5+.
+        Can we use native UPSERTs?
         """
-        return self._version >= 90500
+        return True
 
     def is_deadlock(self, error):
         if isinstance(error, self.module.DatabaseError):

synapse/storage/search.py

@@ -40,8 +40,6 @@ class SearchStore(BackgroundUpdateStore):
 
     EVENT_SEARCH_UPDATE_NAME = "event_search"
     EVENT_SEARCH_ORDER_UPDATE_NAME = "event_search_order"
-    EVENT_SEARCH_USE_GIST_POSTGRES_NAME = "event_search_postgres_gist"
-    EVENT_SEARCH_USE_GIN_POSTGRES_NAME = "event_search_postgres_gin"
 
     def __init__(self, db_conn, hs):
         super(SearchStore, self).__init__(db_conn, hs)
@@ -56,17 +54,6 @@ def __init__(self, db_conn, hs):
             self.EVENT_SEARCH_ORDER_UPDATE_NAME, self._background_reindex_search_order
         )
 
-        # we used to have a background update to turn the GIN index into a
-        # GIST one; we no longer do that (obviously) because we actually want
-        # a GIN index. However, it's possible that some people might still have
-        # the background update queued, so we register a handler to clear the
-        # background update.
-        self.register_noop_background_update(self.EVENT_SEARCH_USE_GIST_POSTGRES_NAME)
-
-        self.register_background_update_handler(
-            self.EVENT_SEARCH_USE_GIN_POSTGRES_NAME, self._background_reindex_gin_search
-        )
-
     @defer.inlineCallbacks
     def _background_reindex_search(self, progress, batch_size):
         # we work through the events table from highest stream id to lowest
@@ -168,49 +155,6 @@ def reindex_search_txn(txn):
 
         defer.returnValue(result)
 
-    @defer.inlineCallbacks
-    def _background_reindex_gin_search(self, progress, batch_size):
-        """This handles old synapses which used GIST indexes, if any;
-        converting them back to be GIN as per the actual schema.
-        """
-
-        def create_index(conn):
-            conn.rollback()
-
-            # we have to set autocommit, because postgres refuses to
-            # CREATE INDEX CONCURRENTLY without it.
-            conn.set_session(autocommit=True)
-
-            try:
-                c = conn.cursor()
-
-                # if we skipped the conversion to GIST, we may already/still
-                # have an event_search_fts_idx; unfortunately postgres 9.4
-                # doesn't support CREATE INDEX IF EXISTS so we just catch the
-                # exception and ignore it.
-                import psycopg2
-
-                try:
-                    c.execute(
-                        "CREATE INDEX CONCURRENTLY event_search_fts_idx"
-                        " ON event_search USING GIN (vector)"
-                    )
-                except psycopg2.ProgrammingError as e:
-                    logger.warn(
-                        "Ignoring error %r when trying to switch from GIST to GIN", e
-                    )
-
-                # we should now be able to delete the GIST index.
-                c.execute("DROP INDEX IF EXISTS event_search_fts_idx_gist")
-            finally:
-                conn.set_session(autocommit=False)
-
-        if isinstance(self.database_engine, PostgresEngine):
-            yield self.runWithConnection(create_index)
-
-        yield self._end_background_update(self.EVENT_SEARCH_USE_GIN_POSTGRES_NAME)
-        defer.returnValue(1)
-
     @defer.inlineCallbacks
     def _background_reindex_search_order(self, progress, batch_size):
         target_min_stream_id = progress["target_min_stream_id_inclusive"]
@@ -341,29 +285,7 @@ def store_search_entries_txn(self, txn, entries):
                 for entry in entries
             )
 
-            # inserts to a GIN index are normally batched up into a pending
-            # list, and then all committed together once the list gets to a
-            # certain size. The trouble with that is that postgres (pre-9.5)
-            # uses work_mem to determine the length of the list, and work_mem
-            # is typically very large.
-            #
-            # We therefore reduce work_mem while we do the insert.
-            #
-            # (postgres 9.5 uses the separate gin_pending_list_limit setting,
-            # so doesn't suffer the same problem, but changing work_mem will
-            # be harmless)
-            #
-            # Note that we don't need to worry about restoring it on
-            # exception, because exceptions will cause the transaction to be
-            # rolled back, including the effects of the SET command.
-            #
-            # Also: we use SET rather than SET LOCAL because there's lots of
-            # other stuff going on in this transaction, which want to have the
-            # normal work_mem setting.
-
-            txn.execute("SET work_mem='256kB'")
             txn.executemany(sql, args)
-            txn.execute("RESET work_mem")
 
         elif isinstance(self.database_engine, Sqlite3Engine):
             sql = (