ci: Nightly fixes (2026-08-18) (#38314)

Based on https://buildkite.com/materialize/nightly/builds/18109 and
nightly failures on my PRs yesterday

Test run: https://buildkite.com/materialize/nightly/builds/18111

Author: def-

misc/python/materialize/parallel_workload/action.py

@@ -389,12 +389,19 @@ def errors_to_ignore(self, exe: Executor) -> list[str]:
             # created after the backup vanish while still being tracked.
             # "invalid database" is the CREATE SCHEMA wording for a database
             # that vanished the same way (CreateSchemaAction does not lock it).
+            # For ZeroDowntimeDeploy the not-yet-promoted environmentd opens
+            # the catalog in savepoint mode, a snapshot frozen at its boot
+            # time, so a worker that lands on it does not see anything the
+            # leader created since. Name resolution runs before the read-only
+            # check, so such a statement reports the missing object rather than
+            # "cannot write in read-only mode".
             result.extend(
                 [
                     "unknown catalog item",
                     "unknown schema",
                     "unknown database",
                     "invalid database",
+                    "unknown network policy",
                 ]
             )
         if exe.db.scenario == Scenario.Rename:
@@ -5965,11 +5972,6 @@ def errors_to_ignore(self, exe: Executor) -> list[str]:
             # The policy is installed as a default somewhere (should not happen,
             # we never install ours, but be safe).
             "cannot be dropped",
-            # Another worker dropped the same policy first. The error carries
-            # the raw name ('netpol-N', not the quoted form), so DROP resolves
-            # the name correctly. This is a concurrency race, not the ALTER
-            # NETWORK POLICY quoted-name bug.
-            "unknown network policy",
         ] + super().errors_to_ignore(exe)
 
     def run(self, exe: Executor) -> bool:

src/mysql-util/src/partition.rs

@@ -392,6 +392,7 @@ mod tests {
     }
 
     #[mz_ore::test(tokio::test)]
+    #[cfg_attr(miri, ignore)] // too slow
     async fn splits_evenly_across_workers() -> Result<(), MySqlError> {
         let mut db = MockDb::new(keys(200_000));
         let count = u64::cast_from(db.keys.len());
@@ -463,6 +464,7 @@ mod tests {
     }
 
     #[mz_ore::test(tokio::test)]
+    #[cfg_attr(miri, ignore)] // too slow
     async fn probe_budget_bounds_requests() -> Result<(), MySqlError> {
         // Confirms baseline over 200 requests.
         let mut db = MockDb::new(keys(200_000));

test/scalability/mzcompose.py

@@ -69,7 +69,6 @@
     scalability_framework_result_storage,
 )
 from materialize.test_analytics.test_analytics_db import TestAnalyticsDb
-from materialize.ui import UIError
 from materialize.util import YesNoOnce, all_subclasses
 from materialize.version_ancestor_overrides import (
     ANCESTOR_OVERRIDES_FOR_SCALABILITY_REGRESSIONS,
@@ -340,12 +339,14 @@ def get_baseline_and_other_endpoints(
                 )
 
                 if resolved_target is None:
-                    if specified_target == regression_against_target:
-                        raise UIError(
-                            f"Could not resolve --regression-against={specified_target} "
-                            "to an image tag, so no regression comparison can be performed"
-                        )
-
+                    # No comparable ancestor image exists. The common case is
+                    # an accepted regression that landed on main and demands a
+                    # baseline newer than any published release (see
+                    # ANCESTOR_OVERRIDES_FOR_SCALABILITY_REGRESSIONS); only
+                    # release images share HEAD's build profile, so there is
+                    # nothing fair to measure against until that release ships.
+                    # Drop the target and keep benchmarking the remaining ones,
+                    # so the measurements still reach test analytics.
                     print(
                         f"Skipping target {specified_target}: no ancestor image could be resolved"
                     )
@@ -424,7 +425,7 @@ def report_assessment(regression_assessment: RegressionAssessment):
     print("+++ Assessment of regressions")
 
     if not regression_assessment.has_comparison_target():
-        print("No comparison was performed because not baseline was specified")
+        print("No comparison was performed because there is no baseline endpoint")
         return
 
     assert regression_assessment.baseline_endpoint is not None