isolate concurrent CI runs on shared real buckets via GITHUB_RUN_ID

Three concurrent Test (1.26, 25.8) jobs (master push + PR runs) shared the
same remote path backup/{cluster}/{shard}/{version} and deleted each other's
fixed-name test_rebase_gcs_* backups mid-test, see
https://github.com/Altinity/clickhouse-backup/actions/runs/29487098583/job/87610918468

- append GITHUB_RUN_ID to the {version} macro (writeMacrosVersionXML), so
  every integration config path/object_disk_path is unique per workflow run
- testflows: storage_prefix includes GITHUB_RUN_ID (pid is not unique
  across runners)
- disk_cos endpoint gets ${HOSTNAME} like disk_gcs already had
- ReadBackupMetadataRemote returns an explicit error for Broken backups
  instead of an empty BackupMetadata that produced the confusing
  "required backup ... has data_format=" error

Author: Slach

pkg/backup/download.go

@@ -1831,6 +1831,12 @@ func (b *Backuper) ReadBackupMetadataRemote(ctx context.Context, backupName stri
 	}
 	for _, backup := range backupList {
 		if backup.BackupName == backupName {
+			// a Broken placeholder contains only BackupName, returning it would produce
+			// confusing downstream errors about empty data_format/tables (e.g. a backup
+			// being deleted or uploaded by a concurrent process)
+			if backup.Broken != "" {
+				return nil, errors.Errorf("%s is %s on remote storage", backupName, backup.Broken)
+			}
 			return &backup.BackupMetadata, nil
 		}
 	}

test/integration/configs/dynamic_settings.sh

@@ -217,7 +217,7 @@ cat <<EOT > /etc/clickhouse-server/config.d/storage_configuration_cos.xml
     <disks>
       <disk_cos>
         ${COS_DISK_TYPE}
-        <endpoint>https://clickhouse-backup-1336113806.cos.na-ashburn.myqcloud.com/disk_cos/{cluster}/{shard}/</endpoint>
+        <endpoint>https://clickhouse-backup-1336113806.cos.na-ashburn.myqcloud.com/disk_cos/${HOSTNAME}/{cluster}/{shard}/</endpoint>
         <access_key_id>${QA_TENCENT_SECRET_ID}</access_key_id>
         <secret_access_key>${QA_TENCENT_SECRET_KEY}</secret_access_key>
         <send_metadata>false</send_metadata>

test/integration/containers.go

@@ -1093,8 +1093,14 @@ func (tc *TestContainers) clickHouseBinds(curDir, configsDir string) []string {
 // dynamic_settings.sh runs only in advanced mode, while old versions (1.x, 19.x) need the
 // isolation too: with the macro missing `{version}` stays literal, so all simple-mode matrix
 // jobs collide on the same remote path and delete each other's fixed-name backups.
+// GITHUB_RUN_ID is appended because the version alone doesn't isolate CONCURRENT workflow
+// runs (master push + PR) executing the same matrix job: they share the remote path and
+// delete/overwrite each other's fixed-name backups mid-test.
 func (tc *TestContainers) writeMacrosVersionXML() (string, error) {
 	version := strings.ReplaceAll(getEnvDefault("CLICKHOUSE_VERSION", "26.3"), ".", "_")
+	if runID := os.Getenv("GITHUB_RUN_ID"); runID != "" {
+		version += "-" + runID
+	}
 	content := fmt.Sprintf("<yandex>\n  <macros>\n    <version>%s</version>\n  </macros>\n</yandex>\n", version)
 	fPath := filepath.Join(os.TempDir(), fmt.Sprintf("clickhouse-backup-macros-version-env%d.xml", tc.envID))
 	if err := os.WriteFile(fPath, []byte(content), 0644); err != nil {

test/testflows/clickhouse_backup/regression.py

@@ -80,7 +80,9 @@ def regression(self, local, stress=False, fips=True, fips_godebug="only"):
     config_dir = f"{cwd}/configs/backup_{os.getpid()}"
     shutil.copytree(base_config_dir, config_dir, dirs_exist_ok=True)
 
-    storage_prefix = f"testflows_{os.getpid()}"
+    # GITHUB_RUN_ID isolates concurrent CI workflow runs sharing the same real bucket
+    # (pid alone is not unique across runners), pid isolates parallel runs on one host
+    storage_prefix = f"testflows_{os.environ.get('GITHUB_RUN_ID', 'local')}_{os.getpid()}"
     origin_path = f"{config_dir}/config.yml.origin"
     config_path = f"{config_dir}/config.yml"
     with open(origin_path) as f: