Add Valkey 9.1.0 support

Author: VGoshev

README.md

@@ -9,7 +9,7 @@ It uses a patched valkey version to make a cluster or sentinel-like setup less p
 
 ## Limitations and requirements
 
-* Patched valkey (patches for valkey 9.0 are included in this repo)
+* Patched valkey (patches for valkey 9.1 are included in this repo)
 * ZooKeeper as DCS
 * Single valkey instance per host
 * In clustered setup each shard must have it's own DCS prefix

valkey_patches/0001_Add_replication_pause.patch

@@ -1,8 +1,8 @@
 diff --git a/src/config.c b/src/config.c
-index d0158b2c4..f788aef6f 100644
+index fcbaee7e9..5c58b971f 100644
 --- a/src/config.c
 +++ b/src/config.c
-@@ -2588,6 +2588,25 @@ static int updateExtendedRedisCompat(const char **err) {
+@@ -2641,6 +2641,25 @@ static int updateExtendedRedisCompat(const char **err) {
      return 1;
  }
 
@@ -28,16 +28,16 @@ index d0158b2c4..f788aef6f 100644
  static int updateSighandlerEnabled(const char **err) {
      UNUSED(err);
      if (server.crashlog_enabled)
-@@ -3241,6 +3260,7 @@ standardConfig static_configs[] = {
-     createBoolConfig("hide-user-data-from-log", NULL, MODIFIABLE_CONFIG, server.hide_user_data_from_log, 1, NULL, NULL),
+@@ -3312,6 +3331,7 @@ standardConfig static_configs[] = {
      createBoolConfig("lua-enable-insecure-api", "lua-enable-deprecated-api", MODIFIABLE_CONFIG | HIDDEN_CONFIG | PROTECTED_CONFIG, server.lua_enable_insecure_api, 0, NULL, updateLuaEnableInsecureApi),
      createBoolConfig("import-mode", NULL, DEBUG_CONFIG | MODIFIABLE_CONFIG, server.import_mode, 0, NULL, NULL),
+     createBoolConfig("io-threads-always-active", NULL, MODIFIABLE_CONFIG | HIDDEN_CONFIG, server.io_threads_always_active, 0, NULL, NULL),
 +    createBoolConfig("repl-paused", NULL, MODIFIABLE_CONFIG, server.repl_paused, 0, NULL, updateReplPaused),
 
      /* String Configs */
      createStringConfig("aclfile", NULL, IMMUTABLE_CONFIG, ALLOW_EMPTY_STRING, server.acl_filename, "", NULL, NULL),
 diff --git a/src/replication.c b/src/replication.c
-index 82ee9450f..25726c542 100644
+index 13cbc4183..c30560b3d 100644
 --- a/src/replication.c
 +++ b/src/replication.c
 @@ -59,7 +59,6 @@ void replicationResurrectProvisionalPrimary(void);
@@ -48,7 +48,7 @@ index 82ee9450f..25726c542 100644
  void replicationSteadyStateInit(void);
  void dualChannelSetupMainConnForPsync(connection *conn);
  void dualChannelSyncHandleRdbLoadCompletion(void);
-@@ -1069,7 +1068,7 @@ void syncCommand(client *c) {
+@@ -1115,7 +1114,7 @@ void syncCommand(client *c) {
 
      /* Refuse SYNC requests if we are a replica but the link with our primary
       * is not ok... */
@@ -57,7 +57,7 @@ index 82ee9450f..25726c542 100644
          addReplyError(c, "-NOMASTERLINK Can't SYNC while not connected with my master");
          return;
      }
-@@ -5150,7 +5149,7 @@ void replicationCron(void) {
+@@ -5284,7 +5283,7 @@ void replicationCron(void) {
      }
 
      /* Check if we should connect to a PRIMARY */
@@ -67,80 +67,26 @@ index 82ee9450f..25726c542 100644
          connectWithPrimary();
      }
 diff --git a/src/server.h b/src/server.h
-index 25f01a31e..95d31758e 100644
+index 351b1e43c..c8e03be25 100644
 --- a/src/server.h
 +++ b/src/server.h
-@@ -2117,6 +2117,7 @@ struct valkeyServer {
+@@ -2205,6 +2205,7 @@ struct valkeyServer {
      /* Synchronous replication. */
      list *clients_waiting_acks; /* Clients waiting in WAIT or WAITAOF. */
      int get_ack_from_replicas;  /* If true we send REPLCONF GETACK. */
 +    int repl_paused;            /* If true we don't try to connect to master */
      /* Limits */
      unsigned int maxclients;                    /* Max number of simultaneous clients */
      unsigned long long maxmemory;               /* Max number of memory bytes to use */
-@@ -3116,6 +3117,8 @@ void updateFailoverStatus(void);
+@@ -3240,6 +3241,8 @@ void updateFailoverStatus(void);
  void abortFailover(const char *err);
  const char *getFailoverStateString(void);
  sds getReplicaPortString(void);
 +int cancelReplicationHandshake(int reconnect);
 +int replicaIsInHandshakeState(void);
  int sendCurrentOffsetToReplica(client *replica);
+ int replicaRdbVersion(client *replica);
  void addRdbReplicaToPsyncWait(client *replica);
- void initClientReplicationData(client *c);
-diff --git a/tests/cluster/tests/99-yandex-cloud-patches.tcl b/tests/cluster/tests/99-yandex-cloud-patches.tcl
-new file mode 100644
-index 000000000..6d0c1007b
---- /dev/null
-+++ b/tests/cluster/tests/99-yandex-cloud-patches.tcl
-@@ -0,0 +1,48 @@
-+# Test Yandex Cloud patches on cluster
-+
-+source "../tests/includes/init-tests.tcl"
-+
-+proc kill_clustered_redis {id} {
-+    set pid [get_instance_attrib redis $id pid]
-+
-+    stop_instance $pid
-+    set_instance_attrib redis $id pid -1
-+    set_instance_attrib redis $id link you_tried_to_talk_with_killed_instance
-+
-+    # Remove the PID from the list of pids to kill at exit.
-+    set ::pids [lsearch -all -inline -not -exact $::pids $pid]
-+}
-+
-+test "Create a 2 node cluster (1 master and 1 replica)" {
-+    create_cluster 1 1
-+}
-+
-+test "Cluster is up" {
-+    assert_cluster_state ok
-+}
-+
-+test "Instance #1 synced with the master" {
-+    wait_for_condition 1000 50 {
-+        [RI 1 master_link_status] eq {up}
-+    } else {
-+        fail "Instance #1 master link status is not up"
-+    }
-+}
-+
-+test "Replication pause on instance #1 works" {
-+    assert {[R 1 config set repl-paused yes] eq {OK}}
-+    wait_for_condition 1000 50 {
-+        [RI 1 master_link_status] eq {down}
-+    } else {
-+        fail "Instance #1 master link status is not down"
-+    }
-+}
-+
-+test "Replication resume on instance #1 works" {
-+    assert {[R 1 config set repl-paused no] eq {OK}}
-+    wait_for_condition 1000 50 {
-+        [RI 1 master_link_status] eq {up}
-+    } else {
-+        fail "Instance #1 master link status is not up"
-+    }
-+}
 diff --git a/tests/integration/yandex-cloud-patches.tcl b/tests/integration/yandex-cloud-patches.tcl
 new file mode 100644
 index 000000000..a2c9bb949
@@ -203,6 +149,49 @@ index 000000000..a2c9bb949
 +        }
 +    }
 +}
--- 
-2.51.1
-
+diff --git a/tests/unit/cluster/99-yandex-cloud-patches.tcl b/tests/unit/cluster/99-yandex-cloud-patches.tcl
+new file mode 100644
+index 000000000..ca4d2f5ec
+--- /dev/null
++++ b/tests/unit/cluster/99-yandex-cloud-patches.tcl
+@@ -0,0 +1,40 @@
++# Test Yandex Cloud patches on cluster
++
++start_cluster 1 2 {tags {external:skip cluster}} {
++
++test "Cluster is up" {
++    wait_for_cluster_state ok
++}
++
++test "Instance 0 is master and 1 is replica" {
++    assert {[s 0 role] eq {master}}
++    assert {[s -1 role] eq {slave}}
++}
++
++test "Instance 1 synced with the master" {
++    wait_for_condition 1000 50 {
++        [s -1 master_link_status] eq {up}
++    } else {
++        fail "Instance 1 master link status is not up"
++    }
++}
++
++test "Replication pause on instance 1 works" {
++    assert {[R 1 config set repl-paused yes] eq {OK}}
++    wait_for_condition 1000 50 {
++        [s -1 master_link_status] eq {down}
++    } else {
++        fail "Instance 1 master link status is not down"
++    }
++}
++
++test "Replication resume on instance 1 works" {
++    assert {[R 1 config set repl-paused no] eq {OK}}
++    wait_for_condition 1000 50 {
++        [s -1 master_link_status] eq {up}
++    } else {
++        fail "Instance 1 master link status is not up"
++    }
++}
++
++}

valkey_patches/0002_Allow_explicit_cluster_replication_cascades.patch

@@ -1,8 +1,8 @@
 diff --git a/src/cluster_legacy.c b/src/cluster_legacy.c
-index ed4768bfe..cc6fb7d1d 100644
+index 63cfc13a0..10adddf5c 100644
 --- a/src/cluster_legacy.c
 +++ b/src/cluster_legacy.c
-@@ -7733,7 +7733,7 @@ int clusterCommandSpecial(client *c) {
+@@ -8120,7 +8120,7 @@ int clusterCommandSpecial(client *c) {
          }
 
          /* Can't replicate a replica. */
@@ -11,38 +11,24 @@ index ed4768bfe..cc6fb7d1d 100644
              addReplyError(c, "I can only replicate a master, not a replica.");
              return 1;
          }
-diff --git a/tests/cluster/tests/99-yandex-cloud-patches.tcl b/tests/cluster/tests/99-yandex-cloud-patches.tcl
-index 6d0c1007b..04b644128 100644
---- a/tests/cluster/tests/99-yandex-cloud-patches.tcl
-+++ b/tests/cluster/tests/99-yandex-cloud-patches.tcl
-@@ -13,8 +13,8 @@ proc kill_clustered_redis {id} {
-     set ::pids [lsearch -all -inline -not -exact $::pids $pid]
- }
- 
--test "Create a 2 node cluster (1 master and 1 replica)" {
--    create_cluster 1 1
-+test "Create a 3 node cluster (1 master and 2 replicas)" {
-+    create_cluster 1 2
- }
- 
- test "Cluster is up" {
-@@ -38,6 +38,16 @@ test "Replication pause on instance #1 works" {
+diff --git a/tests/unit/cluster/99-yandex-cloud-patches.tcl b/tests/unit/cluster/99-yandex-cloud-patches.tcl
+index ca4d2f5ec..bb7844a92 100644
+--- a/tests/unit/cluster/99-yandex-cloud-patches.tcl
++++ b/tests/unit/cluster/99-yandex-cloud-patches.tcl
+@@ -28,6 +28,16 @@ test "Replication pause on instance 1 works" {
      }
  }
 
 +test "Replication cascade with paused instance works" {
 +    assert {[R 2 config set cluster-slave-no-failover yes] eq {OK}}
-+    assert {[R 2 cluster replicate [R 1 CLUSTER MYID]] eq {OK}}
++    assert {[R 2 cluster replicate [R 1 cluster myid]] eq {OK}}
 +    wait_for_condition 1000 50 {
-+        [RI 2 master_link_status] eq {up}
++        [s -2 master_link_status] eq {up}
 +    } else {
-+        fail "Instance #2 master link status is not up"
++        fail "Instance 2 master link status is not up"
 +    }
 +}
 +
- test "Replication resume on instance #1 works" {
+ test "Replication resume on instance 1 works" {
      assert {[R 1 config set repl-paused no] eq {OK}}
      wait_for_condition 1000 50 {
--- 
-2.51.1
-

valkey_patches/0003_Add_offline_mode.patch

@@ -1,8 +1,8 @@
 diff --git a/src/config.c b/src/config.c
-index f788aef6f..aafa0616a 100644
+index 5c58b971f..0d852e730 100644
 --- a/src/config.c
 +++ b/src/config.c
-@@ -623,6 +623,8 @@ void loadServerConfigFromString(sds config) {
+@@ -631,6 +631,8 @@ void loadServerConfigFromString(sds config) {
      if (server.hz < CONFIG_MIN_HZ) server.hz = CONFIG_MIN_HZ;
      if (server.hz > CONFIG_MAX_HZ) server.hz = CONFIG_MAX_HZ;
 
@@ -11,7 +11,7 @@ index f788aef6f..aafa0616a 100644
      sdsfreesplitres(lines, totlines);
      reading_config_file = 0;
      return;
-@@ -3206,6 +3208,31 @@ static int applyClientMaxMemoryUsage(const char **err) {
+@@ -3276,6 +3278,31 @@ static int isValidDbHashSeed(sds val, const char **err) {
      return 1;
  }
 
@@ -43,7 +43,7 @@ index f788aef6f..aafa0616a 100644
  standardConfig static_configs[] = {
      /* Bool configs */
      createBoolConfig("rdbchecksum", NULL, IMMUTABLE_CONFIG, server.rdb_checksum, 1, NULL, NULL),
-@@ -3461,6 +3488,7 @@ standardConfig static_configs[] = {
+@@ -3536,6 +3563,7 @@ standardConfig static_configs[] = {
      createSpecialConfig("rdma-bind", NULL, MODIFIABLE_CONFIG | MULTI_ARG_CONFIG, setConfigRdmaBindOption, getConfigRdmaBindOption, rewriteConfigRdmaBindOption, applyRdmaBind),
      createSpecialConfig("replicaof", "slaveof", IMMUTABLE_CONFIG | MULTI_ARG_CONFIG, setConfigReplicaOfOption, getConfigReplicaOfOption, rewriteConfigReplicaOfOption, NULL),
      createSpecialConfig("latency-tracking-info-percentiles", NULL, MODIFIABLE_CONFIG | MULTI_ARG_CONFIG, setConfigLatencyTrackingInfoPercentilesOutputOption, getConfigLatencyTrackingInfoPercentilesOutputOption, rewriteConfigLatencyTrackingInfoPercentilesOutputOption, NULL),
@@ -52,10 +52,10 @@ index f788aef6f..aafa0616a 100644
      /* NULL Terminator, this is dropped when we convert to the runtime array. */
      {NULL},
 diff --git a/src/server.c b/src/server.c
-index 46a20d1ae..860a3cb41 100644
+index e018a6182..7594a672e 100644
 --- a/src/server.c
 +++ b/src/server.c
-@@ -2664,9 +2664,15 @@ int listenToPort(connListener *sfd) {
+@@ -2717,9 +2717,15 @@ int listenToPort(connListener *sfd) {
      int j;
      int port = sfd->port;
      char **bindaddr = sfd->bindaddr;
@@ -72,7 +72,7 @@ index 46a20d1ae..860a3cb41 100644
 
      for (j = 0; j < sfd->bindaddr_count; j++) {
          char *addr = bindaddr[j];
-@@ -6865,8 +6871,8 @@ void dismissMemoryInChild(void) {
+@@ -7176,8 +7182,8 @@ void dismissMemoryInChild(void) {
      /* madvise(MADV_DONTNEED) may not work if Transparent Huge Pages is enabled. */
      if (server.thp_enabled) return;
 
@@ -83,20 +83,11 @@ index 46a20d1ae..860a3cb41 100644
  #if defined(USE_JEMALLOC) && defined(__linux__)
      listIter li;
      listNode *ln;
-@@ -7311,7 +7317,7 @@ __attribute__((weak)) int main(int argc, char **argv) {
-     }
-     if (server.sentinel_mode) sentinelCheckConfigFile();
- 
--        /* Do system checks */
-+    /* Do system checks */
- #ifdef __linux__
-     linuxMemoryWarnings();
-     sds err_msg = NULL;
 diff --git a/src/server.h b/src/server.h
-index 95d31758e..a82eabd26 100644
+index c8e03be25..d941665ea 100644
 --- a/src/server.h
 +++ b/src/server.h
-@@ -2118,6 +2118,9 @@ struct valkeyServer {
+@@ -2206,6 +2206,9 @@ struct valkeyServer {
      list *clients_waiting_acks; /* Clients waiting in WAIT or WAITAOF. */
      int get_ack_from_replicas;  /* If true we send REPLCONF GETACK. */
      int repl_paused;            /* If true we don't try to connect to master */
@@ -135,6 +126,3 @@ index 000000000..b8c3ba453
 +        }
 +    }
 +}
--- 
-2.51.1
-