[incubator-pegasus] branch master updated: fix(update_replication_factor#12): update the mutation_2pc_min_replica_count base the max count of table (#1035)

[la...@apache.org]

This is an automated email from the ASF dual-hosted git repository.

laiyingchun pushed a commit to branch master
in repository https://gitbox.apache.org/repos/asf/incubator-pegasus.git


The following commit(s) were added to refs/heads/master by this push:
     new 502fd2026 fix(update_replication_factor#12): update the mutation_2pc_min_replica_count base the max count of table (#1035)
502fd2026 is described below

commit 502fd20262b437e1b64a7eff561803f15f37e990
Author: Jiashuo <js...@live.com>
AuthorDate: Tue Jul 12 15:32:48 2022 +0800

    fix(update_replication_factor#12): update the mutation_2pc_min_replica_count base the max count of table (#1035)
---
 .gitignore                             |  8 ++++++++
 rdsn/src/common/replication_common.cpp | 13 ++++++++++++-
 rdsn/src/common/replication_common.h   |  1 +
 rdsn/src/meta/partition_guardian.cpp   |  8 ++++----
 rdsn/src/meta/partition_guardian.h     |  1 -
 rdsn/src/meta/server_state.cpp         |  6 ++++--
 rdsn/src/replica/replica_2pc.cpp       |  4 ++--
 rdsn/src/replica/replica_config.cpp    |  2 +-
 scripts/format_files.sh                |  5 +++--
 9 files changed, 35 insertions(+), 13 deletions(-)

diff --git a/.gitignore b/.gitignore
index be06ef135..0e1294ea9 100644
--- a/.gitignore
+++ b/.gitignore
@@ -261,3 +261,11 @@ package-lock.json
 
 #go
 go-client/bin
+
+# rdsn
+rdsn/builder
+rdsn/thirdparty/build
+rdsn/thirdparty/src
+rdsn/thirdparty/output
+rdsn/cmake-build-debug
+rdsn/test_reports
diff --git a/rdsn/src/common/replication_common.cpp b/rdsn/src/common/replication_common.cpp
index 3f3afdb59..46c2e6fd3 100644
--- a/rdsn/src/common/replication_common.cpp
+++ b/rdsn/src/common/replication_common.cpp
@@ -230,7 +230,8 @@ void replication_options::initialize()
         "replication",
         "mutation_2pc_min_replica_count",
         mutation_2pc_min_replica_count,
-        "minimum number of alive replicas under which write is allowed");
+        "minimum number of alive replicas under which write is allowed. it's valid if larger than "
+        "0, otherwise, the final value is based on app_max_replica_count");
 
     group_check_disabled = dsn_config_get_value_bool("replication",
                                                      "group_check_disabled",
@@ -421,6 +422,16 @@ void replication_options::sanity_check()
             staleness_for_commit);
 }
 
+int32_t replication_options::app_mutation_2pc_min_replica_count(int32_t app_max_replica_count) const
+{
+    dcheck_gt(app_max_replica_count, 0);
+    if (mutation_2pc_min_replica_count > 0) { //  >0 means use the user config
+        return mutation_2pc_min_replica_count;
+    } else { // otherwise, the value based on the table max_replica_count
+        return app_max_replica_count <= 2 ? 1 : app_max_replica_count / 2 + 1;
+    }
+}
+
 /*static*/ bool replica_helper::remove_node(::dsn::rpc_address node,
                                             /*inout*/ std::vector<::dsn::rpc_address> &nodeList)
 {
diff --git a/rdsn/src/common/replication_common.h b/rdsn/src/common/replication_common.h
index 28e800020..813ea771b 100644
--- a/rdsn/src/common/replication_common.h
+++ b/rdsn/src/common/replication_common.h
@@ -117,6 +117,7 @@ public:
     ~replication_options();
 
     void initialize();
+    int32_t app_mutation_2pc_min_replica_count(int32_t app_max_replica_count) const;
     static bool get_data_dir_and_tag(const std::string &config_dirs_str,
                                      const std::string &default_dir,
                                      const std::string &app_name,
diff --git a/rdsn/src/meta/partition_guardian.cpp b/rdsn/src/meta/partition_guardian.cpp
index ce343b5fb..f2eb22e3e 100644
--- a/rdsn/src/meta/partition_guardian.cpp
+++ b/rdsn/src/meta/partition_guardian.cpp
@@ -24,13 +24,11 @@ namespace replication {
 partition_guardian::partition_guardian(meta_service *svc) : _svc(svc)
 {
     if (svc != nullptr) {
-        _mutation_2pc_min_replica_count = svc->get_options().mutation_2pc_min_replica_count;
         _replica_assign_delay_ms_for_dropouts =
             svc->get_meta_options()._lb_opts.replica_assign_delay_ms_for_dropouts;
         config_context::MAX_REPLICA_COUNT_IN_GRROUP =
             svc->get_meta_options()._lb_opts.max_replicas_in_group;
     } else {
-        _mutation_2pc_min_replica_count = 0;
         _replica_assign_delay_ms_for_dropouts = 0;
     }
 
@@ -477,8 +475,10 @@ pc_status partition_guardian::on_missing_secondary(meta_view &view, const dsn::g
 
     configuration_proposal_action action;
     bool is_emergency = false;
-    if (cc.config_owner->max_replica_count > _mutation_2pc_min_replica_count &&
-        replica_count(pc) < _mutation_2pc_min_replica_count) {
+    if (cc.config_owner->max_replica_count >
+            _svc->get_options().app_mutation_2pc_min_replica_count(pc.max_replica_count) &&
+        replica_count(pc) <
+            _svc->get_options().app_mutation_2pc_min_replica_count(pc.max_replica_count)) {
         // ATTENTION:
         // when max_replica_count == 2, even if there is only 1 replica alive now, we will still
         // wait for replica_assign_delay_ms_for_dropouts before recover the second replica.
diff --git a/rdsn/src/meta/partition_guardian.h b/rdsn/src/meta/partition_guardian.h
index 1d20551f2..73efef250 100644
--- a/rdsn/src/meta/partition_guardian.h
+++ b/rdsn/src/meta/partition_guardian.h
@@ -91,7 +91,6 @@ private:
     // ]
     dsn_handle_t _ctrl_assign_secondary_black_list = nullptr;
 
-    int32_t _mutation_2pc_min_replica_count;
     dsn_handle_t _ctrl_assign_delay_ms = nullptr;
     uint64_t _replica_assign_delay_ms_for_dropouts;
 
diff --git a/rdsn/src/meta/server_state.cpp b/rdsn/src/meta/server_state.cpp
index f852504cf..7be63bfd1 100644
--- a/rdsn/src/meta/server_state.cpp
+++ b/rdsn/src/meta/server_state.cpp
@@ -1446,7 +1446,8 @@ void server_state::update_configuration_locally(
     partition_configuration &old_cfg = app.partitions[gpid.get_partition_index()];
     partition_configuration &new_cfg = config_request->config;
 
-    int min_2pc_count = _meta_svc->get_options().mutation_2pc_min_replica_count;
+    int min_2pc_count =
+        _meta_svc->get_options().app_mutation_2pc_min_replica_count(app.max_replica_count);
     health_status old_health_status = partition_health_status(old_cfg, min_2pc_count);
     health_status new_health_status = partition_health_status(new_cfg, min_2pc_count);
 
@@ -2375,8 +2376,9 @@ void server_state::update_partition_perf_counter()
 {
     int counters[HS_MAX_VALUE];
     ::memset(counters, 0, sizeof(counters));
-    int min_2pc_count = _meta_svc->get_options().mutation_2pc_min_replica_count;
     auto func = [&](const std::shared_ptr<app_state> &app) {
+        int min_2pc_count =
+            _meta_svc->get_options().app_mutation_2pc_min_replica_count(app->max_replica_count);
         for (unsigned int i = 0; i != app->partition_count; ++i) {
             health_status st = partition_health_status(app->partitions[i], min_2pc_count);
             counters[st]++;
diff --git a/rdsn/src/replica/replica_2pc.cpp b/rdsn/src/replica/replica_2pc.cpp
index 314ded57b..124bf5662 100644
--- a/rdsn/src/replica/replica_2pc.cpp
+++ b/rdsn/src/replica/replica_2pc.cpp
@@ -144,7 +144,7 @@ void replica::on_client_write(dsn::message_ex *request, bool ignore_throttling)
     }
 
     if (static_cast<int>(_primary_states.membership.secondaries.size()) + 1 <
-        _options->mutation_2pc_min_replica_count) {
+        _options->app_mutation_2pc_min_replica_count(_app_info.max_replica_count)) {
         response_client_write(request, ERR_NOT_ENOUGH_MEMBER);
         return;
     }
@@ -223,7 +223,7 @@ void replica::init_prepare(mutation_ptr &mu, bool reconciliation, bool pop_all_c
     // for reconciliation, we should ensure every prepared mutation to be committed
     // please refer to PacificA paper
     if (static_cast<int>(_primary_states.membership.secondaries.size()) + 1 <
-            _options->mutation_2pc_min_replica_count &&
+            _options->app_mutation_2pc_min_replica_count(_app_info.max_replica_count) &&
         !reconciliation) {
         err = ERR_NOT_ENOUGH_MEMBER;
         goto ErrOut;
diff --git a/rdsn/src/replica/replica_config.cpp b/rdsn/src/replica/replica_config.cpp
index b79cf082f..4ea4727d9 100644
--- a/rdsn/src/replica/replica_config.cpp
+++ b/rdsn/src/replica/replica_config.cpp
@@ -1068,7 +1068,7 @@ bool replica::update_local_configuration(const replica_configuration &config,
         }
 
         if (_primary_states.membership.secondaries.size() + 1 <
-            _options->mutation_2pc_min_replica_count) {
+            _options->app_mutation_2pc_min_replica_count(_app_info.max_replica_count)) {
             std::vector<mutation_ptr> queued;
             _primary_states.write_queue.clear(queued);
             for (auto &m : queued) {
diff --git a/scripts/format_files.sh b/scripts/format_files.sh
index af477a13a..28ed3b309 100755
--- a/scripts/format_files.sh
+++ b/scripts/format_files.sh
@@ -22,17 +22,18 @@ cd $root_dir
 
 linenoise=./src/shell/linenoise
 sds=./src/shell/sds
+thirdparty=./rdsn/thirdparty
 
 if [ $# -eq 0 ]; then
   echo "formating all .h/.cpp files in $root_dir ..."
-  find . -type f -not \( -wholename "$linenoise/*" -o -wholename "$sds/*" \) \
+  find . -type f -not \( -wholename "$linenoise/*" -o -wholename "$sds/*" -o -wholename "$thirdparty/*" \) \
       -regextype posix-egrep -regex ".*\.(cpp|h)" | xargs clang-format-3.9 -i -style=file
 elif [ $1 = "-h" ]; then
   echo "USAGE: ./format-files.sh [<relative_path>] -- format .h/.cpp files in $root_dir/relative_path"
   echo "       ./format-files.sh means format all .h/.cpp files in $root_dir"
 else
   echo "formating all .h/.cpp files in $root_dir/$1 ..."
-  find ./$1 -type f -not \( -wholename "$linenoise/*" -o -wholename "$sds/*" \) \
+  find ./$1 -type f -not \( -wholename "$linenoise/*" -o -wholename "$sds/*" -o -wholename "$thirdparty/*" \) \
       -regextype posix-egrep -regex ".*\.(cpp|h)" | xargs clang-format-3.9 -i -style=file
 fi
 


---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscribe@pegasus.apache.org
For additional commands, e-mail: commits-help@pegasus.apache.org