mirrors/scylladb
1
0
Fork 0
mirror of https://github.com/scylladb/scylladb.git synced 2026-05-28 10:41:12 +00:00
Code Issues Packages Projects Releases Wiki Activity

schema: Fix race in schema version recalculation leading to stale schema version in gossip

Browse Source 
Migration manager installs several feature change listeners:

    if (this_shard_id() == 0) {
        _feature_listeners.push_back(_feat.cluster_supports_view_virtual_columns().when_enabled(update_schema));
        _feature_listeners.push_back(_feat.cluster_supports_digest_insensitive_to_expiry().when_enabled(update_schema));
        _feature_listeners.push_back(_feat.cluster_supports_cdc().when_enabled(update_schema));
        _feature_listeners.push_back(_feat.cluster_supports_per_table_partitioners().when_enabled(update_schema));
    }

They will call update_schema_version_and_announce() when features are enabled, which does this:

    return update_schema_version(proxy, features).then([] (utils::UUID uuid) {
        return announce_schema_version(uuid);
    });

So it first updates the schema version and then publishes it via
gossip in announce_schema_version(). It is possible that the
announce_schema_version() part of the first schema change will be
deferred and will execute after the other four calls to
update_schema_version_and_announce(). It will install the old schema
version in gossip instead of the more recent one.

The fix is to serialize schema digest calculation and publishing.

Fixes #7200

(cherry picked from commit 1a57d641d1)
This commit is contained in:
Tomasz Grabiec
2020-09-10 00:28:12 +02:00
parent f8c7605657
commit e99a0c7b89
3 changed files with 16 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
db/schema_tables.cc
View File

@@ -822,6 +822,14 @@ future<> merge_schema(distributed<service::storage_proxy>& proxy, gms::feature_s
});
}
future<> recalculate_schema_version(distributed<service::storage_proxy>& proxy, gms::feature_service& feat) {
return merge_lock().then([&proxy, &feat] {
return update_schema_version_and_announce(proxy, feat.cluster_schema_features());
}).finally([] {
return merge_unlock();
});
}
future<> merge_schema(distributed<service::storage_proxy>& proxy, std::vector<mutation> mutations, bool do_flush)
{
return merge_lock().then([&proxy, mutations = std::move(mutations), do_flush] () mutable {

7
db/schema_tables.hh
View File

@@ -170,6 +170,13 @@ future<> merge_schema(distributed<service::storage_proxy>& proxy, gms::feature_s
future<> merge_schema(distributed<service::storage_proxy>& proxy, std::vector<mutation> mutations, bool do_flush);
// Recalculates the local schema version and publishes it in gossip.
//
// It is safe to call concurrently with recalculate_schema_version() and merge_schema() in which case it
// is guaranteed that the schema version we end up with after all calls will reflect the most recent state
// of feature_service and schema tables.
future<> recalculate_schema_version(distributed<service::storage_proxy>& proxy, gms::feature_service& feat);
future<std::set<sstring>> merge_keyspaces(distributed<service::storage_proxy>& proxy, schema_result&& before, schema_result&& after);
std::vector<mutation> make_create_keyspace_mutations(lw_shared_ptr<keyspace_metadata> keyspace, api::timestamp_type timestamp, bool with_tables_and_types_and_functions = true);

2
service/migration_manager.cc
View File

@@ -92,7 +92,7 @@ void migration_manager::init_messaging_service()
//FIXME: future discarded.
(void)with_gate(_background_tasks, [this] {
mlogger.debug("features changed, recalculating schema version");
return update_schema_version_and_announce(get_storage_proxy(), _feat.cluster_schema_features());
return db::schema_tables::recalculate_schema_version(get_storage_proxy(), _feat);
});
};