Merge pull request #36196 from vespa-engine/hmusum/stop-using-delete-tenant-feature-flags

Stop using feature flags related to deleting tenants

Author: hmusum

configserver/src/main/java/com/yahoo/vespa/config/server/tenant/TenantRepository.java

@@ -37,10 +37,7 @@
 import com.yahoo.vespa.curator.Curator;
 import com.yahoo.vespa.curator.transaction.CuratorOperations;
 import com.yahoo.vespa.curator.transaction.CuratorTransaction;
-import com.yahoo.vespa.flags.BooleanFlag;
 import com.yahoo.vespa.flags.FlagSource;
-import com.yahoo.vespa.flags.Flags;
-import com.yahoo.vespa.flags.LongFlag;
 import org.apache.curator.framework.CuratorFramework;
 import org.apache.curator.framework.recipes.cache.PathChildrenCacheEvent;
 import org.apache.curator.framework.state.ConnectionState;
@@ -112,7 +109,6 @@ public class TenantRepository {
     private final FileDistributionFactory fileDistributionFactory;
     private final ExecutorService deployHelperExecutor;
     private final FlagSource flagSource;
-    private final BooleanFlag softDeleteTenantFlag;
     private final HostProvisionerProvider hostProvisionerProvider;
     private final ConfigserverConfig configserverConfig;
     private final ConfigServerDB configServerDB;
@@ -128,7 +124,6 @@ public class TenantRepository {
     private final List<EndpointCertificateSecretStore> endpointCertificateSecretStores;
     private final OnnxModelCost onnxModelCost;
     private final InheritableApplications inheritableApplications;
-    private final LongFlag deleteIdleTenantSecondsFlag;
 
     /**
      * Creates a new tenant repository
@@ -204,7 +199,6 @@ public TenantRepository(HostRegistry hostRegistry,
         this.zkSessionWatcherExecutor = zkSessionWatcherExecutor;
         this.fileDistributionFactory = fileDistributionFactory;
         this.flagSource = flagSource;
-        this.softDeleteTenantFlag = Flags.SOFT_DELETE_TENANT.bindTo(flagSource);
         this.hostProvisionerProvider = hostProvisionerProvider;
         this.configServerDB = configServerDB;
         this.zone = zone;
@@ -215,7 +209,6 @@ public TenantRepository(HostRegistry hostRegistry,
         this.tenantListener = tenantListener;
         this.zookeeperServerConfig = zookeeperServerConfig;
         this.endpointCertificateSecretStores = endpointCertificateSecretStores;
-        this.deleteIdleTenantSecondsFlag = Flags.DELETE_IDLE_TENANT_SECONDS.bindTo(flagSource);
         // This we should control with a feature flag.
         this.deployHelperExecutor = createModelBuilderExecutor();
         this.onnxModelCost = onnxModelCost;
@@ -270,7 +263,7 @@ private TenantMetaData createMetaData(Tenant tenant) {
         Instant now = tenant.getSessionRepository().clock().instant();
         TenantMetaData metadata = getTenantMetaData(tenant);
         Instant deployTime = metadata.lastDeployTimestamp();
-        if (deployTime.equals(Instant.EPOCH) || deleteIdleTenantSecondsFlag.with(tenant.getName()).value() < 0)
+        if (deployTime.equals(Instant.EPOCH))
             deployTime = now;
         Instant createdTime = metadata.createdTimestamp();
         if (createdTime.equals(Instant.EPOCH))
@@ -489,22 +482,17 @@ public void deleteTenant(TenantName name) {
 
             notifyRemovedTenant(name);
             tenant.close();
-            if (softDeleteTenantFlag.with(name).value()) {
-                // Because each config server has a PathDirectoryCache on the `sessions` and `applications` children:
-                //  1. Once the first config server (say cfg1) reaches this point, the caches on cfg2-3 will recreate
-                //    `sessions` and `applications` immediately after they are deleted by this tryDelete(), likely
-                //    failing this deletion.
-                //  2. Once the next config server (say cfg2) also reaches this point, the caches on cfg3 will recreate
-                //     `sessions` and `applications`, possibly failing this deletion.
-                //  3. Once the last config server (cfg3) reaches this point, the delete should succeed.
-                if (curator.tryDelete(tenant.getPath())) {
-                    log.log(Level.INFO, "Deleted tenant " + name);
-                } else {
-                    log.log(Level.INFO, "Deleted tenant " + name + " (" + tenant.getPath() + " to be removed by other cfgs)");
-                }
+            // Because each config server has a PathDirectoryCache on the `sessions` and `applications` children:
+            //  1. Once the first config server (say cfg1) reaches this point, the caches on cfg2-3 will recreate
+            //    `sessions` and `applications` immediately after they are deleted by this tryDelete(), likely
+            //    failing this deletion.
+            //  2. Once the next config server (say cfg2) also reaches this point, the caches on cfg3 will recreate
+            //     `sessions` and `applications`, possibly failing this deletion.
+            //  3. Once the last config server (cfg3) reaches this point, the delete should succeed.
+            if (curator.tryDelete(tenant.getPath())) {
+                log.log(Level.INFO, "Deleted tenant " + name);
             } else {
-                curator.delete(tenant.getPath());
-                log.log(Level.INFO, "Deleted tenant '" + name + "'");
+                log.log(Level.INFO, "Deleted tenant " + name + " (" + tenant.getPath() + " to be removed by other cfgs)");
             }
         }
     }
@@ -661,11 +649,7 @@ public Set<TenantName> deleteUnusedTenants(Duration ttlForUnusedTenant, Instant
                 .filter(tenantName -> activeApplications(tenantName).isEmpty())
                 .filter(tenantName -> !tenantName.equals(TenantName.defaultName())) // Not allowed to remove 'default' tenant
                 .filter(tenantName -> !tenantName.equals(HOSTED_VESPA_TENANT)) // Not allowed to remove 'hosted-vespa' tenant
-                .filter(tenantName -> {
-                    long ttlSeconds = deleteIdleTenantSecondsFlag.with(tenantName).value();
-                    Duration ttl = ttlSeconds < 0 ? ttlForUnusedTenant : Duration.ofSeconds(ttlSeconds);
-                    return getTenantMetaData(getTenant(tenantName)).lastDeployTimestamp().isBefore(now.minus(ttl));
-                })
+                .filter(tenantName -> getTenantMetaData(getTenant(tenantName)).lastDeployTimestamp().isBefore(now.minus(ttlForUnusedTenant)))
                 .peek(this::deleteTenant)
                 .collect(Collectors.toSet());
     }

flags/src/main/java/com/yahoo/vespa/flags/Flags.java

@@ -66,23 +66,6 @@ public class Flags {
             "Takes effect on next deployment of the application",
             INSTANCE_ID, VESPA_VERSION);
 
-    // TODO(2026-02-20): Remove once this has rolled out, and the overrides have been removed.
-    public static final UnboundLongFlag DELETE_IDLE_TENANT_SECONDS = defineLongFlag(
-            "delete-idle-tenant-seconds", 604800,
-            List.of("hakonhall"), "2026-02-03", "2026-04-03",
-            "If >=0, then (A) the last deploy time is not updated on config server bootstrap, " +
-            "and (B) an idle tenant will be deleted after this many seconds (default 604800 = 1 week).",
-            "(A) takes effect on cfg bootstrap, (B) on next tick of TenantsMaintainer",
-            TENANT_ID);
-
-    // TODO(2026-02-20): Remove once this has rolled out, and the overrides have been removed.
-    public static final UnboundBooleanFlag SOFT_DELETE_TENANT = defineFeatureFlag(
-            "soft-delete-tenant", true,
-            List.of("hakonhall"), "2026-01-20", "2026-04-20",
-            "When deleting /config/v2/tenants/TENANT recursively - whether to give up (true) or retry (false) on NotEmptyException",
-            "Takes effect immediately",
-            TENANT_ID);
-
     public static final UnboundBooleanFlag LOCKED_GCP_PROVISION = defineFeatureFlag(
             "locked-gcp-provision", true,
             List.of("hakonhall"), "2025-08-05", "2026-04-15",