Merge branch 'main' into erik/history-size-consumption-metric

test_runner/fixtures/endpoint/http.py

@@ -1,6 +1,7 @@
 from __future__ import annotations
 
 import urllib.parse
+from enum import StrEnum
 from typing import TYPE_CHECKING, final
 
 import requests
@@ -9,11 +10,23 @@ from requests.auth import AuthBase
 from typing_extensions import override
 
 from fixtures.log_helper import log
+from fixtures.utils import wait_until
 
 if TYPE_CHECKING:
     from requests import PreparedRequest
 
 
+COMPUTE_AUDIENCE = "compute"
+"""
+The value to place in the `aud` claim.
+"""
+
+
+@final
+class ComputeClaimsScope(StrEnum):
+    ADMIN = "admin"
+
+
 @final
 class BearerAuth(AuthBase):
     """
@@ -50,6 +63,35 @@ class EndpointHttpClient(requests.Session):
         res.raise_for_status()
         return res.json()
 
+    def prewarm_lfc_status(self) -> dict[str, str]:
+        res = self.get(f"http://localhost:{self.external_port}/lfc/prewarm")
+        res.raise_for_status()
+        json: dict[str, str] = res.json()
+        return json
+
+    def prewarm_lfc(self):
+        self.post(f"http://localhost:{self.external_port}/lfc/prewarm").raise_for_status()
+
+        def prewarmed():
+            json = self.prewarm_lfc_status()
+            status, err = json["status"], json.get("error")
+            assert status == "completed", f"{status}, error {err}"
+
+        wait_until(prewarmed)
+
+    def offload_lfc(self):
+        url = f"http://localhost:{self.external_port}/lfc/offload"
+        self.post(url).raise_for_status()
+
+        def offloaded():
+            res = self.get(url)
+            res.raise_for_status()
+            json = res.json()
+            status, err = json["status"], json.get("error")
+            assert status == "completed", f"{status}, error {err}"
+
+        wait_until(offloaded)
+
     def database_schema(self, database: str):
         res = self.get(
             f"http://localhost:{self.external_port}/database_schema?database={urllib.parse.quote(database, safe='')}",

test_runner/fixtures/neon_cli.py

@@ -21,6 +21,7 @@ if TYPE_CHECKING:
         Any,
     )
 
+    from fixtures.endpoint.http import ComputeClaimsScope
     from fixtures.pg_version import PgVersion
 
 
@@ -535,12 +536,16 @@ class NeonLocalCli(AbstractNeonCli):
         res.check_returncode()
         return res
 
-    def endpoint_generate_jwt(self, endpoint_id: str) -> str:
+    def endpoint_generate_jwt(
+        self, endpoint_id: str, scope: ComputeClaimsScope | None = None
+    ) -> str:
         """
         Generate a JWT for making requests to the endpoint's external HTTP
         server.
         """
         args = ["endpoint", "generate-jwt", endpoint_id]
+        if scope:
+            args += ["--scope", str(scope)]
 
         cmd = self.raw_cli(args)
         cmd.check_returncode()

test_runner/fixtures/neon_fixtures.py

@@ -51,7 +51,7 @@ from fixtures.common_types import (
     TimelineId,
 )
 from fixtures.compute_migrations import NUM_COMPUTE_MIGRATIONS
-from fixtures.endpoint.http import EndpointHttpClient
+from fixtures.endpoint.http import ComputeClaimsScope, EndpointHttpClient
 from fixtures.log_helper import log
 from fixtures.metrics import Metrics, MetricsGetter, parse_metrics
 from fixtures.neon_cli import NeonLocalCli, Pagectl
@@ -1185,7 +1185,9 @@ class NeonEnv:
             "broker": {},
             "safekeepers": [],
             "pageservers": [],
-            "endpoint_storage": {"port": self.port_distributor.get_port()},
+            "endpoint_storage": {
+                "listen_addr": f"127.0.0.1:{self.port_distributor.get_port()}",
+            },
             "generate_local_ssl_certs": self.generate_local_ssl_certs,
         }
 
@@ -1279,7 +1281,8 @@ class NeonEnv:
                 )
 
             tenant_config = ps_cfg.setdefault("tenant_config", {})
-            tenant_config["rel_size_v2_enabled"] = True  # Enable relsize_v2 by default in tests
+            # This feature is pending rollout.
+            # tenant_config["rel_size_v2_enabled"] = True
 
             if self.pageserver_remote_storage is not None:
                 ps_cfg["remote_storage"] = remote_storage_to_toml_dict(
@@ -4217,7 +4220,7 @@ class Endpoint(PgProtocol, LogUtils):
 
         self.config(config_lines)
 
-        self.__jwt = self.env.neon_cli.endpoint_generate_jwt(self.endpoint_id)
+        self.__jwt = self.generate_jwt()
 
         return self
 
@@ -4264,6 +4267,14 @@ class Endpoint(PgProtocol, LogUtils):
 
         return self
 
+    def generate_jwt(self, scope: ComputeClaimsScope | None = None) -> str:
+        """
+        Generate a JWT for making requests to the endpoint's external HTTP
+        server.
+        """
+        assert self.endpoint_id is not None
+        return self.env.neon_cli.endpoint_generate_jwt(self.endpoint_id, scope)
+
     def endpoint_path(self) -> Path:
         """Path to endpoint directory"""
         assert self.endpoint_id

test_runner/performance/test_cumulative_statistics_persistence.py

@@ -1,4 +1,5 @@
 import math  # Add this import
+import os
 import time
 import traceback
 from pathlib import Path
@@ -87,7 +88,10 @@ def test_cumulative_statistics_persistence(
     - insert additional tuples that by itself are not enough to trigger auto-vacuum but in combination with the previous tuples are
     - verify that autovacuum is triggered by the combination of tuples inserted before and after endpoint suspension
     """
-    project = neon_api.create_project(pg_version)
+    project = neon_api.create_project(
+        pg_version,
+        f"Test cumulative statistics persistence, GITHUB_RUN_ID={os.getenv('GITHUB_RUN_ID')}",
+    )
     project_id = project["project"]["id"]
     neon_api.wait_for_operation_to_finish(project_id)
     endpoint_id = project["endpoints"][0]["id"]

test_runner/performance/test_physical_replication.py

@@ -62,7 +62,9 @@ def test_ro_replica_lag(
 
     pgbench_duration = f"-T{test_duration_min * 60 * 2}"
 
-    project = neon_api.create_project(pg_version)
+    project = neon_api.create_project(
+        pg_version, f"Test readonly replica lag, GITHUB_RUN_ID={os.getenv('GITHUB_RUN_ID')}"
+    )
     project_id = project["project"]["id"]
     log.info("Project ID: %s", project_id)
     log.info("Primary endpoint ID: %s", project["endpoints"][0]["id"])
@@ -195,7 +197,9 @@ def test_replication_start_stop(
     pgbench_duration = f"-T{2**num_replicas * configuration_test_time_sec}"
     error_occurred = False
 
-    project = neon_api.create_project(pg_version)
+    project = neon_api.create_project(
+        pg_version, f"Test replication start stop, GITHUB_RUN_ID={os.getenv('GITHUB_RUN_ID')}"
+    )
     project_id = project["project"]["id"]
     log.info("Project ID: %s", project_id)
     log.info("Primary endpoint ID: %s", project["endpoints"][0]["id"])

test_runner/random_ops/test_random_ops.py

@@ -206,7 +206,7 @@ class NeonProject:
         self.neon_api = neon_api
         self.pg_bin = pg_bin
         proj = self.neon_api.create_project(
-            pg_version, f"Automatic random API test {os.getenv('GITHUB_RUN_ID')}"
+            pg_version, f"Automatic random API test GITHUB_RUN_ID={os.getenv('GITHUB_RUN_ID')}"
         )
         self.id: str = proj["project"]["id"]
         self.name: str = proj["project"]["name"]

test_runner/regress/test_attach_tenant_config.py

@@ -186,7 +186,7 @@ def test_fully_custom_config(positive_env: NeonEnv):
             "type": "interpreted",
             "args": {"format": "bincode", "compression": {"zstd": {"level": 1}}},
         },
-        "rel_size_v2_enabled": False,  # test suite enables it by default as of https://github.com/neondatabase/neon/issues/11081, so, custom config means disabling it
+        "rel_size_v2_enabled": True,
         "gc_compaction_enabled": True,
         "gc_compaction_verification": False,
         "gc_compaction_initial_threshold_kb": 1024000,

test_runner/regress/test_compaction.py

@@ -202,6 +202,8 @@ def test_pageserver_gc_compaction_preempt(
     env = neon_env_builder.init_start(initial_tenant_conf=conf)
 
     env.pageserver.allowed_errors.append(".*The timeline or pageserver is shutting down.*")
+    env.pageserver.allowed_errors.append(".*flush task cancelled.*")
+    env.pageserver.allowed_errors.append(".*failed to pipe.*")
 
     tenant_id = env.initial_tenant
     timeline_id = env.initial_timeline
@@ -229,7 +231,7 @@ def test_pageserver_gc_compaction_preempt(
 
 
 @skip_in_debug_build("only run with release build")
-@pytest.mark.timeout(600)  # This test is slow with sanitizers enabled, especially on ARM
+@pytest.mark.timeout(900)  # This test is slow with sanitizers enabled, especially on ARM
 @pytest.mark.parametrize(
     "with_branches",
     ["with_branches", "no_branches"],

test_runner/regress/test_compute_catalog.py

@@ -544,3 +544,69 @@ def test_drop_role_with_table_privileges_from_non_neon_superuser(neon_simple_env
         )
         role = cursor.fetchone()
         assert role is None
+
+
+def test_db_with_custom_settings(neon_simple_env: NeonEnv):
+    """
+    Test that compute_ctl can work with databases that have some custom settings.
+    For example, role=some_other_role, default_transaction_read_only=on,
+    search_path=non_public_schema, statement_timeout=1 (1ms).
+    """
+    env = neon_simple_env
+
+    endpoint = env.endpoints.create_start("main")
+
+    TEST_ROLE = "some_other_role"
+    TEST_DB = "db_with_custom_settings"
+    TEST_SCHEMA = "non_public_schema"
+
+    endpoint.respec_deep(
+        **{
+            "spec": {
+                "skip_pg_catalog_updates": False,
+                "cluster": {
+                    "databases": [
+                        {
+                            "name": TEST_DB,
+                            "owner": TEST_ROLE,
+                        }
+                    ],
+                    "roles": [
+                        {
+                            "name": TEST_ROLE,
+                        }
+                    ],
+                },
+            }
+        }
+    )
+
+    endpoint.reconfigure()
+
+    with endpoint.cursor(dbname=TEST_DB) as cursor:
+        cursor.execute(f"CREATE SCHEMA {TEST_SCHEMA}")
+        cursor.execute(f"ALTER DATABASE {TEST_DB} SET role = {TEST_ROLE}")
+        cursor.execute(f"ALTER DATABASE {TEST_DB} SET default_transaction_read_only = on")
+        cursor.execute(f"ALTER DATABASE {TEST_DB} SET search_path = {TEST_SCHEMA}")
+        cursor.execute(f"ALTER DATABASE {TEST_DB} SET statement_timeout = 1")
+
+    with endpoint.cursor(dbname=TEST_DB) as cursor:
+        cursor.execute("SELECT current_role")
+        role = cursor.fetchone()
+        assert role is not None
+        assert role[0] == TEST_ROLE
+
+        cursor.execute("SHOW default_transaction_read_only")
+        default_transaction_read_only = cursor.fetchone()
+        assert default_transaction_read_only is not None
+        assert default_transaction_read_only[0] == "on"
+
+        cursor.execute("SHOW search_path")
+        search_path = cursor.fetchone()
+        assert search_path is not None
+        assert search_path[0] == TEST_SCHEMA
+
+        # Do not check statement_timeout, because we force it to 2min
+        # in `endpoint.cursor()` fixture.
+
+    endpoint.reconfigure()

test_runner/regress/test_compute_http.py

@@ -0,0 +1,78 @@
+from __future__ import annotations
+
+from http.client import FORBIDDEN, UNAUTHORIZED
+from typing import TYPE_CHECKING
+
+import jwt
+import pytest
+from fixtures.endpoint.http import COMPUTE_AUDIENCE, ComputeClaimsScope, EndpointHttpClient
+from fixtures.utils import run_only_on_default_postgres
+from requests import RequestException
+
+if TYPE_CHECKING:
+    from fixtures.neon_fixtures import NeonEnv
+
+
+@run_only_on_default_postgres("The code path being tested is not dependent on Postgres version")
+def test_compute_no_scope_claim(neon_simple_env: NeonEnv):
+    """
+    Test that if the JWT scope is not admin and no compute_id is specified,
+    the external HTTP server returns a 403 Forbidden error.
+    """
+    env = neon_simple_env
+
+    endpoint = env.endpoints.create_start("main")
+
+    # Encode nothing in the token
+    token = jwt.encode({}, env.auth_keys.priv, algorithm="EdDSA")
+
+    # Create an admin-scoped HTTP client
+    client = EndpointHttpClient(
+        external_port=endpoint.external_http_port,
+        internal_port=endpoint.internal_http_port,
+        jwt=token,
+    )
+
+    try:
+        client.status()
+        pytest.fail("Exception should have been raised")
+    except RequestException as e:
+        assert e.response is not None
+        assert e.response.status_code == FORBIDDEN
+
+
+@pytest.mark.parametrize(
+    "audience",
+    (COMPUTE_AUDIENCE, "invalid", None),
+    ids=["with_audience", "with_invalid_audience", "without_audience"],
+)
+@run_only_on_default_postgres("The code path being tested is not dependent on Postgres version")
+def test_compute_admin_scope_claim(neon_simple_env: NeonEnv, audience: str | None):
+    """
+    Test that an admin-scoped JWT can access the compute's external HTTP server
+    without the compute_id being specified in the claims.
+    """
+    env = neon_simple_env
+
+    endpoint = env.endpoints.create_start("main")
+
+    data: dict[str, str | list[str]] = {"scope": str(ComputeClaimsScope.ADMIN)}
+    if audience:
+        data["aud"] = [audience]
+
+    token = jwt.encode(data, env.auth_keys.priv, algorithm="EdDSA")
+
+    # Create an admin-scoped HTTP client
+    client = EndpointHttpClient(
+        external_port=endpoint.external_http_port,
+        internal_port=endpoint.internal_http_port,
+        jwt=token,
+    )
+
+    try:
+        client.status()
+        if audience != COMPUTE_AUDIENCE:
+            pytest.fail("Exception should have been raised")
+    except RequestException as e:
+        assert e.response is not None
+        assert e.response.status_code == UNAUTHORIZED

test_runner/regress/test_endpoint_storage.py

@@ -4,10 +4,12 @@ import pytest
 from aiohttp import ClientSession
 from fixtures.log_helper import log
 from fixtures.neon_fixtures import NeonEnv
+from fixtures.utils import run_only_on_default_postgres
 from jwcrypto import jwk, jwt
 
 
 @pytest.mark.asyncio
+@run_only_on_default_postgres("test doesn't use postgres")
 async def test_endpoint_storage_insert_retrieve_delete(neon_simple_env: NeonEnv):
     """
     Inserts, retrieves, and deletes test file using a JWT token
@@ -35,7 +37,6 @@ async def test_endpoint_storage_insert_retrieve_delete(neon_simple_env: NeonEnv)
     key = f"http://{base_url}/{tenant_id}/{timeline_id}/{endpoint_id}/key"
     headers = {"Authorization": f"Bearer {token}"}
     log.info(f"cache key url {key}")
-    log.info(f"token {token}")
 
     async with ClientSession(headers=headers) as session:
         async with session.get(key) as res:

test_runner/regress/test_gist.py

@@ -0,0 +1,28 @@
+from __future__ import annotations
+
+from typing import TYPE_CHECKING
+
+if TYPE_CHECKING:
+    from fixtures.neon_fixtures import NeonEnv
+
+
+#
+# Test unlogged build for GIST index
+#
+def test_gist(neon_simple_env: NeonEnv):
+    env = neon_simple_env
+    endpoint = env.endpoints.create_start("main")
+    con = endpoint.connect()
+    cur = con.cursor()
+    iterations = 100
+
+    for _ in range(iterations):
+        cur.execute(
+            "CREATE TABLE pvactst (i INT, a INT[], p POINT) with (autovacuum_enabled = off)"
+        )
+        cur.execute(
+            "INSERT INTO pvactst SELECT i, array[1,2,3], point(i, i+1) FROM generate_series(1,1000) i"
+        )
+        cur.execute("CREATE INDEX gist_pvactst ON pvactst USING gist (p)")
+        cur.execute("VACUUM pvactst")
+        cur.execute("DROP TABLE pvactst")

test_runner/regress/test_lfc_prewarm.py

@@ -1,11 +1,24 @@
 import random
 import threading
 import time
+from enum import Enum
 
 import pytest
+from fixtures.endpoint.http import EndpointHttpClient
 from fixtures.log_helper import log
 from fixtures.neon_fixtures import NeonEnv
 from fixtures.utils import USE_LFC
+from prometheus_client.parser import text_string_to_metric_families as prom_parse_impl
+
+
+class LfcQueryMethod(Enum):
+    COMPUTE_CTL = False
+    POSTGRES = True
+
+
+PREWARM_LABEL = "compute_ctl_lfc_prewarm_requests_total"
+OFFLOAD_LABEL = "compute_ctl_lfc_offload_requests_total"
+QUERY_OPTIONS = LfcQueryMethod.POSTGRES, LfcQueryMethod.COMPUTE_CTL
 
 
 def check_pinned_entries(cur):
@@ -19,11 +32,20 @@ def check_pinned_entries(cur):
     assert n_pinned == 0
 
 
+def prom_parse(client: EndpointHttpClient) -> dict[str, float]:
+    return {
+        sample.name: sample.value
+        for family in prom_parse_impl(client.metrics())
+        for sample in family.samples
+        if sample.name in (PREWARM_LABEL, OFFLOAD_LABEL)
+    }
+
+
 @pytest.mark.skipif(not USE_LFC, reason="LFC is disabled, skipping")
-def test_lfc_prewarm(neon_simple_env: NeonEnv):
+@pytest.mark.parametrize("query", QUERY_OPTIONS, ids=["postgres", "compute-ctl"])
+def test_lfc_prewarm(neon_simple_env: NeonEnv, query: LfcQueryMethod):
     env = neon_simple_env
     n_records = 1000000
-
     endpoint = env.endpoints.create_start(
         branch_name="main",
         config_lines=[
@@ -34,30 +56,57 @@ def test_lfc_prewarm(neon_simple_env: NeonEnv):
             "neon.file_cache_prewarm_limit=1000",
         ],
     )
-    conn = endpoint.connect()
-    cur = conn.cursor()
-    cur.execute("create extension neon version '1.6'")
-    cur.execute("create table t(pk integer primary key, payload text default repeat('?', 128))")
-    cur.execute(f"insert into t (pk) values (generate_series(1,{n_records}))")
-    cur.execute("select get_local_cache_state()")
-    lfc_state = cur.fetchall()[0][0]
+
+    pg_conn = endpoint.connect()
+    pg_cur = pg_conn.cursor()
+    pg_cur.execute("create extension neon version '1.6'")
+    pg_cur.execute("create database lfc")
+
+    lfc_conn = endpoint.connect(dbname="lfc")
+    lfc_cur = lfc_conn.cursor()
+    log.info(f"Inserting {n_records} rows")
+    lfc_cur.execute("create table t(pk integer primary key, payload text default repeat('?', 128))")
+    lfc_cur.execute(f"insert into t (pk) values (generate_series(1,{n_records}))")
+    log.info(f"Inserted {n_records} rows")
+
+    http_client = endpoint.http_client()
+    if query is LfcQueryMethod.COMPUTE_CTL:
+        status = http_client.prewarm_lfc_status()
+        assert status["status"] == "not_prewarmed"
+        assert "error" not in status
+        http_client.offload_lfc()
+        assert http_client.prewarm_lfc_status()["status"] == "not_prewarmed"
+        assert prom_parse(http_client) == {OFFLOAD_LABEL: 1, PREWARM_LABEL: 0}
+    else:
+        pg_cur.execute("select get_local_cache_state()")
+        lfc_state = pg_cur.fetchall()[0][0]
 
     endpoint.stop()
     endpoint.start()
 
-    conn = endpoint.connect()
-    cur = conn.cursor()
-    time.sleep(1)  # wait until compute_ctl complete downgrade of extension to default version
-    cur.execute("alter extension neon update to '1.6'")
-    cur.execute("select prewarm_local_cache(%s)", (lfc_state,))
+    # wait until compute_ctl completes downgrade of extension to default version
+    time.sleep(1)
+    pg_conn = endpoint.connect()
+    pg_cur = pg_conn.cursor()
+    pg_cur.execute("alter extension neon update to '1.6'")
 
-    cur.execute("select lfc_value from neon_lfc_stats where lfc_key='file_cache_used_pages'")
-    lfc_used_pages = cur.fetchall()[0][0]
+    lfc_conn = endpoint.connect(dbname="lfc")
+    lfc_cur = lfc_conn.cursor()
+
+    if query is LfcQueryMethod.COMPUTE_CTL:
+        http_client.prewarm_lfc()
+    else:
+        pg_cur.execute("select prewarm_local_cache(%s)", (lfc_state,))
+
+    pg_cur.execute("select lfc_value from neon_lfc_stats where lfc_key='file_cache_used_pages'")
+    lfc_used_pages = pg_cur.fetchall()[0][0]
     log.info(f"Used LFC size: {lfc_used_pages}")
-    cur.execute("select * from get_prewarm_info()")
-    prewarm_info = cur.fetchall()[0]
+    pg_cur.execute("select * from get_prewarm_info()")
+    prewarm_info = pg_cur.fetchall()[0]
     log.info(f"Prewarm info: {prewarm_info}")
-    log.info(f"Prewarm progress: {(prewarm_info[1] + prewarm_info[2]) * 100 // prewarm_info[0]}%")
+    total, prewarmed, skipped, _ = prewarm_info
+    progress = (prewarmed + skipped) * 100 // total
+    log.info(f"Prewarm progress: {progress}%")
 
     assert lfc_used_pages > 10000
     assert (
@@ -66,18 +115,23 @@ def test_lfc_prewarm(neon_simple_env: NeonEnv):
         and prewarm_info[0] == prewarm_info[1] + prewarm_info[2]
     )
 
-    cur.execute("select sum(pk) from t")
-    assert cur.fetchall()[0][0] == n_records * (n_records + 1) / 2
+    lfc_cur.execute("select sum(pk) from t")
+    assert lfc_cur.fetchall()[0][0] == n_records * (n_records + 1) / 2
 
-    check_pinned_entries(cur)
+    check_pinned_entries(pg_cur)
+
+    desired = {"status": "completed", "total": total, "prewarmed": prewarmed, "skipped": skipped}
+    if query is LfcQueryMethod.COMPUTE_CTL:
+        assert http_client.prewarm_lfc_status() == desired
+        assert prom_parse(http_client) == {OFFLOAD_LABEL: 0, PREWARM_LABEL: 1}
 
 
 @pytest.mark.skipif(not USE_LFC, reason="LFC is disabled, skipping")
-def test_lfc_prewarm_under_workload(neon_simple_env: NeonEnv):
+@pytest.mark.parametrize("query", QUERY_OPTIONS, ids=["postgres", "compute-ctl"])
+def test_lfc_prewarm_under_workload(neon_simple_env: NeonEnv, query: LfcQueryMethod):
     env = neon_simple_env
     n_records = 10000
     n_threads = 4
-
     endpoint = env.endpoints.create_start(
         branch_name="main",
         config_lines=[
@@ -87,40 +141,58 @@ def test_lfc_prewarm_under_workload(neon_simple_env: NeonEnv):
             "neon.file_cache_prewarm_limit=1000000",
         ],
     )
-    conn = endpoint.connect()
-    cur = conn.cursor()
-    cur.execute("create extension neon version '1.6'")
-    cur.execute(
+
+    pg_conn = endpoint.connect()
+    pg_cur = pg_conn.cursor()
+    pg_cur.execute("create extension neon version '1.6'")
+    pg_cur.execute("CREATE DATABASE lfc")
+
+    lfc_conn = endpoint.connect(dbname="lfc")
+    lfc_cur = lfc_conn.cursor()
+    lfc_cur.execute(
         "create table accounts(id integer primary key, balance bigint default 0, payload text default repeat('?', 1000)) with (fillfactor=10)"
     )
-    cur.execute(f"insert into accounts(id) values (generate_series(1,{n_records}))")
-    cur.execute("select get_local_cache_state()")
-    lfc_state = cur.fetchall()[0][0]
+    log.info(f"Inserting {n_records} rows")
+    lfc_cur.execute(f"insert into accounts(id) values (generate_series(1,{n_records}))")
+    log.info(f"Inserted {n_records} rows")
+
+    http_client = endpoint.http_client()
+    if query is LfcQueryMethod.COMPUTE_CTL:
+        http_client.offload_lfc()
+    else:
+        pg_cur.execute("select get_local_cache_state()")
+        lfc_state = pg_cur.fetchall()[0][0]
 
     running = True
+    n_prewarms = 0
 
     def workload():
-        conn = endpoint.connect()
-        cur = conn.cursor()
+        lfc_conn = endpoint.connect(dbname="lfc")
+        lfc_cur = lfc_conn.cursor()
         n_transfers = 0
         while running:
             src = random.randint(1, n_records)
             dst = random.randint(1, n_records)
-            cur.execute("update accounts set balance=balance-100 where id=%s", (src,))
-            cur.execute("update accounts set balance=balance+100 where id=%s", (dst,))
+            lfc_cur.execute("update accounts set balance=balance-100 where id=%s", (src,))
+            lfc_cur.execute("update accounts set balance=balance+100 where id=%s", (dst,))
             n_transfers += 1
         log.info(f"Number of transfers: {n_transfers}")
 
     def prewarm():
-        conn = endpoint.connect()
-        cur = conn.cursor()
-        n_prewarms = 0
+        pg_conn = endpoint.connect()
+        pg_cur = pg_conn.cursor()
         while running:
-            cur.execute("alter system set neon.file_cache_size_limit='1MB'")
-            cur.execute("select pg_reload_conf()")
-            cur.execute("alter system set neon.file_cache_size_limit='1GB'")
-            cur.execute("select pg_reload_conf()")
-            cur.execute("select prewarm_local_cache(%s)", (lfc_state,))
+            pg_cur.execute("alter system set neon.file_cache_size_limit='1MB'")
+            pg_cur.execute("select pg_reload_conf()")
+            pg_cur.execute("alter system set neon.file_cache_size_limit='1GB'")
+            pg_cur.execute("select pg_reload_conf()")
+
+            if query is LfcQueryMethod.COMPUTE_CTL:
+                http_client.prewarm_lfc()
+            else:
+                pg_cur.execute("select prewarm_local_cache(%s)", (lfc_state,))
+
+            nonlocal n_prewarms
             n_prewarms += 1
         log.info(f"Number of prewarms: {n_prewarms}")
 
@@ -140,8 +212,10 @@ def test_lfc_prewarm_under_workload(neon_simple_env: NeonEnv):
         t.join()
     prewarm_thread.join()
 
-    cur.execute("select sum(balance) from accounts")
-    total_balance = cur.fetchall()[0][0]
+    lfc_cur.execute("select sum(balance) from accounts")
+    total_balance = lfc_cur.fetchall()[0][0]
     assert total_balance == 0
 
-    check_pinned_entries(cur)
+    check_pinned_entries(pg_cur)
+    if query is LfcQueryMethod.COMPUTE_CTL:
+        assert prom_parse(http_client) == {OFFLOAD_LABEL: 1, PREWARM_LABEL: n_prewarms}

test_runner/regress/test_pageserver_metric_collection.py

@@ -506,7 +506,6 @@ class SyntheticSizeVerifier:
 
 PER_METRIC_VERIFIERS = {
     "remote_storage_size": CannotVerifyAnything,
-    "resident_size": CannotVerifyAnything,
     "written_size": WrittenDataVerifier,
     "written_data_bytes_delta": WrittenDataDeltaVerifier,
     "pitr_cutoff": CannotVerifyAnything,

test_runner/regress/test_pg_regress.py

@@ -471,7 +471,7 @@ def test_tx_abort_with_many_relations(
         try:
             # Rollback phase should be fast: this is one WAL record that we should process efficiently
             fut = exec.submit(rollback_and_wait)
-            fut.result(timeout=15)
+            fut.result(timeout=15 if reldir_type == "v1" else 30)
         except:
             exec.shutdown(wait=False, cancel_futures=True)
             raise

test_runner/regress/test_sharding.py

@@ -1334,6 +1334,13 @@ def test_sharding_split_failures(
         tenant_id, timeline_id, shard_count=initial_shard_count, placement_policy='{"Attached":1}'
     )
 
+    # Create bystander tenants with various shard counts. They should not be affected by the aborted
+    # splits. Regression test for https://github.com/neondatabase/cloud/issues/28589.
+    bystanders = {}  # id → shard_count
+    for bystander_shard_count in [1, 2, 4, 8]:
+        id, _ = env.create_tenant(shard_count=bystander_shard_count)
+        bystanders[id] = bystander_shard_count
+
     env.storage_controller.allowed_errors.extend(
         [
             # All split failures log a warning when then enqueue the abort operation
@@ -1394,6 +1401,8 @@ def test_sharding_split_failures(
             locations = ps.http_client().tenant_list_locations()["tenant_shards"]
             for loc in locations:
                 tenant_shard_id = TenantShardId.parse(loc[0])
+                if tenant_shard_id.tenant_id != tenant_id:
+                    continue  # skip bystanders
                 log.info(f"Shard {tenant_shard_id} seen on node {ps.id} in mode {loc[1]['mode']}")
                 assert tenant_shard_id.shard_count == initial_shard_count
                 if loc[1]["mode"] == "Secondary":
@@ -1414,6 +1423,8 @@ def test_sharding_split_failures(
             locations = ps.http_client().tenant_list_locations()["tenant_shards"]
             for loc in locations:
                 tenant_shard_id = TenantShardId.parse(loc[0])
+                if tenant_shard_id.tenant_id != tenant_id:
+                    continue  # skip bystanders
                 log.info(f"Shard {tenant_shard_id} seen on node {ps.id} in mode {loc[1]['mode']}")
                 assert tenant_shard_id.shard_count == split_shard_count
                 if loc[1]["mode"] == "Secondary":
@@ -1496,6 +1507,12 @@ def test_sharding_split_failures(
     # the scheduler reaches an idle state
     env.storage_controller.reconcile_until_idle(timeout_secs=30)
 
+    # Check that all bystanders are still around.
+    for bystander_id, bystander_shard_count in bystanders.items():
+        response = env.storage_controller.tenant_describe(bystander_id)
+        assert TenantId(response["tenant_id"]) == bystander_id
+        assert len(response["shards"]) == bystander_shard_count
+
     env.storage_controller.consistency_check()