config: improve recommended server config (#294)

src/server/config.ini

@@ -134,13 +134,16 @@
 [threadpool.THREAD_POOL_FDS_SERVICE]
   name = fds_service
   worker_count = 8
+  max_input_queue_length = 1024
+  worker_priority = THREAD_xPRIORITY_NORMAL
+  worker_count = 8
 
 [threadpool.THREAD_POOL_COMPACT]
   name = compact
   partitioned = false
-  max_input_queue_length = 128
+  max_input_queue_length = 1024
   worker_priority = THREAD_xPRIORITY_NORMAL
-  worker_count = 4
+  worker_count = 8
 
 [meta_server]
   server_list = %{meta.server.list}
@@ -152,7 +155,7 @@
   meta_state_service_type = meta_state_service_zookeeper
   meta_state_service_parameters =
 
-  node_live_percentage_threshold_for_update = 60
+  node_live_percentage_threshold_for_update = 50
   min_live_node_count_for_unfreeze = 3
   meta_function_level_on_start = steady
   recover_from_replica_server = false
@@ -164,7 +167,7 @@
 
   server_load_balancer_type = greedy_load_balancer
   replica_assign_delay_ms_for_dropouts = 600000
-  max_replicas_in_group = 4
+  max_replicas_in_group = 3
   balancer_in_turn = false
   only_primary_balancer = false
   only_move_primary = false
@@ -175,6 +178,7 @@
   slog_dir = %{slog.dir}
   data_dirs = %{data.dirs}
   data_dirs_black_list_file = %{home.dir}/.pegasus_data_dirs_black_list
+  cluster_name = %{cluster.name}
 
   deny_client_on_start = false
   verbose_client_log_on_start = false
@@ -203,7 +207,7 @@
   gc_disabled = false
   gc_interval_ms = 30000
   gc_memory_replica_interval_ms = 600000
-  gc_disk_error_replica_interval_seconds = 604800
+  gc_disk_error_replica_interval_seconds = 86400
   gc_disk_garbage_replica_interval_seconds = 86400
 
   disk_stat_disabled = false
@@ -248,14 +252,16 @@
 
 [pegasus.server]
   rocksdb_verbose_log = false
-  rocksdb_abnormal_get_time_threshold_ns = 0
-  rocksdb_abnormal_get_size_threshold = 0
-  rocksdb_abnormal_multi_get_time_threshold_ns = 0
-  rocksdb_abnormal_multi_get_size_threshold = 0
-  rocksdb_abnormal_multi_get_iterate_count_threshold = 0
+
+  # get: {100ms,1MB} ; multiGet: {100ms,10MB,1000}
+  rocksdb_abnormal_get_time_threshold_ns = 100000000
+  rocksdb_abnormal_get_size_threshold = 1000000
+  rocksdb_abnormal_multi_get_time_threshold_ns = 100000000
+  rocksdb_abnormal_multi_get_size_threshold = 10000000
+  rocksdb_abnormal_multi_get_iterate_count_threshold = 1000
 
   rocksdb_write_buffer_size = 67108864
-  rocksdb_max_write_buffer_number = 6
+  rocksdb_max_write_buffer_number = 3
   rocksdb_max_background_flushes = 4
   rocksdb_max_background_compactions = 12
   rocksdb_num_levels = 6
@@ -267,10 +273,12 @@
   rocksdb_level0_slowdown_writes_trigger = 30
   rocksdb_level0_stop_writes_trigger = 60
   rocksdb_disable_table_block_cache = false
+  rocksdb_block_cache_capacity = 10737418240
+  rocksdb_block_cache_num_shard_bits = -1
   rocksdb_compression_type = lz4
 
-  checkpoint_reserve_min_count = 3
-  checkpoint_reserve_time_seconds = 3600
+  checkpoint_reserve_min_count = 2
+  checkpoint_reserve_time_seconds = 1800
 
   updating_rocksdb_sstsize_interval_seconds = 600
 
@@ -451,7 +459,7 @@
 
 [task.RPC_RRDB_RRDB_PUT]
   rpc_request_throttling_mode = TM_DELAY
-  rpc_request_delays_milliseconds = 1000, 1000, 1000, 1000, 1000, 10000
+  rpc_request_delays_milliseconds = 50, 50, 50, 50, 50, 100
   is_profile = true
 
 [task.RPC_RRDB_RRDB_PUT_ACK]
@@ -459,7 +467,7 @@
 
 [task.RPC_RRDB_RRDB_MULTI_PUT]
   rpc_request_throttling_mode = TM_DELAY
-  rpc_request_delays_milliseconds = 1000, 1000, 1000, 1000, 1000, 10000
+  rpc_request_delays_milliseconds = 50, 50, 50, 50, 50, 100
   is_profile = true
 
 [task.RPC_RRDB_RRDB_MULTI_PUT_ACK]
@@ -467,7 +475,7 @@
 
 [task.RPC_RRDB_RRDB_REMOVE]
   rpc_request_throttling_mode = TM_DELAY
-  rpc_request_delays_milliseconds = 1000, 1000, 1000, 1000, 1000, 10000
+  rpc_request_delays_milliseconds = 50, 50, 50, 50, 50, 100
   is_profile = true
 
 [task.RPC_RRDB_RRDB_REMOVE_ACK]
@@ -475,7 +483,7 @@
 
 [task.RPC_RRDB_RRDB_MULTI_REMOVE]
   rpc_request_throttling_mode = TM_DELAY
-  rpc_request_delays_milliseconds = 1000, 1000, 1000, 1000, 1000, 10000
+  rpc_request_delays_milliseconds = 50, 50, 50, 50, 50, 100
   is_profile = true
 
 [task.RPC_RRDB_RRDB_MULTI_REMOVE_ACK]
@@ -483,7 +491,7 @@
 
 [task.RPC_RRDB_RRDB_INCR]
   rpc_request_throttling_mode = TM_DELAY
-  rpc_request_delays_milliseconds = 1000, 1000, 1000, 1000, 1000, 10000
+  rpc_request_delays_milliseconds = 50, 50, 50, 50, 50, 100
   is_profile = true
 
 [task.RPC_RRDB_RRDB_INCR_ACK]
@@ -491,7 +499,7 @@
 
 [task.RPC_RRDB_RRDB_CHECK_AND_SET]
   rpc_request_throttling_mode = TM_DELAY
-  rpc_request_delays_milliseconds = 1000, 1000, 1000, 1000, 1000, 10000
+  rpc_request_delays_milliseconds = 50, 50, 50, 50, 50, 100
   is_profile = true
 
 [task.RPC_RRDB_RRDB_CHECK_AND_SET_ACK]
@@ -499,7 +507,7 @@
 
 [task.RPC_RRDB_RRDB_CHECK_AND_MUTATE]
   rpc_request_throttling_mode = TM_DELAY
-  rpc_request_delays_milliseconds = 1000, 1000, 1000, 1000, 1000, 10000
+  rpc_request_delays_milliseconds = 50, 50, 50, 50, 50, 100
   is_profile = true
 
 [task.RPC_RRDB_RRDB_CHECK_AND_MUTATE_ACK]
@@ -507,7 +515,7 @@
 
 [task.RPC_RRDB_RRDB_GET]
   rpc_request_throttling_mode = TM_DELAY
-  rpc_request_delays_milliseconds = 1000, 1000, 1000, 1000, 1000, 10000
+  rpc_request_delays_milliseconds = 50, 50, 50, 50, 50, 100
   is_profile = true
 
 [task.RPC_RRDB_RRDB_GET_ACK]
@@ -515,7 +523,7 @@
 
 [task.RPC_RRDB_RRDB_MULTI_GET]
   rpc_request_throttling_mode = TM_DELAY
-  rpc_request_delays_milliseconds = 1000, 1000, 1000, 1000, 1000, 10000
+  rpc_request_delays_milliseconds = 50, 50, 50, 50, 50, 100
   is_profile = true
 
 [task.RPC_RRDB_RRDB_MULTI_GET_ACK]
@@ -523,7 +531,7 @@
 
 [task.RPC_RRDB_RRDB_SORTKEY_COUNT]
   rpc_request_throttling_mode = TM_DELAY
-  rpc_request_delays_milliseconds = 1000, 1000, 1000, 1000, 1000, 10000
+  rpc_request_delays_milliseconds = 50, 50, 50, 50, 50, 100
   is_profile = true
 
 [task.RPC_RRDB_RRDB_SORTKEY_COUNT_ACK]
@@ -531,7 +539,7 @@
 
 [task.RPC_RRDB_RRDB_TTL]
   rpc_request_throttling_mode = TM_DELAY
-  rpc_request_delays_milliseconds = 1000, 1000, 1000, 1000, 1000, 10000
+  rpc_request_delays_milliseconds = 50, 50, 50, 50, 50, 100
   is_profile = true
 
 [task.RPC_RRDB_RRDB_TTL_ACK]
@@ -539,7 +547,7 @@
 
 [task.RPC_RRDB_RRDB_GET_SCANNER]
   rpc_request_throttling_mode = TM_DELAY
-  rpc_request_delays_milliseconds = 1000, 1000, 1000, 1000, 1000, 10000
+  rpc_request_delays_milliseconds = 50, 50, 50, 50, 50, 100
   is_profile = true
 
 [task.RPC_RRDB_RRDB_GET_SCANNER_ACK]
@@ -547,7 +555,7 @@
 
 [task.RPC_RRDB_RRDB_SCAN]
   rpc_request_throttling_mode = TM_DELAY
-  rpc_request_delays_milliseconds = 1000, 1000, 1000, 1000, 1000, 10000
+  rpc_request_delays_milliseconds = 50, 50, 50, 50, 50, 100
   is_profile = true
 
 [task.RPC_RRDB_RRDB_SCAN_ACK]
@@ -555,7 +563,7 @@
 
 [task.RPC_RRDB_RRDB_CLEAR_SCANNER]
   rpc_request_throttling_mode = TM_DELAY
-  rpc_request_delays_milliseconds = 1000, 1000, 1000, 1000, 1000, 10000
+  rpc_request_delays_milliseconds = 50, 50, 50, 50, 50, 100
   is_profile = true
 
 [task.RPC_RRDB_RRDB_CLEAR_SCANNER_ACK]