Merge pull request #12584 from rystsov/v23.2.x-11947

[v23.2.x] txns: fix list txns failure when it's used before find_coordinator
redpanda-data · Aug 7, 2023 · e8a873c · e8a873c
2 parents 4ddbeec + c339bce
commit e8a873c
Show file tree

Hide file tree

Showing 4 changed files with 61 additions and 6 deletions.
diff --git a/src/v/cluster/tx_registry_frontend.cc b/src/v/cluster/tx_registry_frontend.cc
@@ -61,14 +61,49 @@ tx_registry_frontend::tx_registry_frontend(
       config::shard_local_cfg().metadata_dissemination_retry_delay_ms.value()) {
 }
 
+ss::future<bool> tx_registry_frontend::ensure_tx_topic_exists() {
+    if (_metadata_cache.local().contains(model::tx_manager_nt)) {
+        co_return true;
+    }
+
+    if (!co_await try_create_tx_topic()) {
+        vlog(clusterlog.warn, "failed to create {}", model::tx_manager_nt);
+        co_return false;
+    }
+
+    auto retries = _metadata_dissemination_retries;
+    auto delay_ms = _metadata_dissemination_retry_delay_ms;
+    std::optional<std::string> error;
+    while (!_as.abort_requested() && 0 < retries--) {
+        auto is_cache_filled = _metadata_cache.local().contains(
+          model::tx_manager_nt);
+        if (unlikely(!is_cache_filled)) {
+            error = vformat(
+              fmt::runtime("can't find {} in the metadata_cache cache"),
+              model::tx_manager_nt);
+            vlog(
+              clusterlog.trace,
+              "waiting for {} to fill metadata_cache cache, retries left: {}",
+              model::tx_manager_nt,
+              retries);
+            co_await sleep_abortable(delay_ms, _as);
+            continue;
+        }
+        co_return true;
+    }
+
+    if (error) {
+        vlog(clusterlog.warn, "{}", error.value());
+    }
+
+    co_return false;
+}
+
 ss::future<find_coordinator_reply> tx_registry_frontend::find_coordinator(
   kafka::transactional_id tid, model::timeout_clock::duration timeout) {
     if (!_metadata_cache.local().contains(model::tx_manager_nt)) {
         if (!co_await try_create_tx_topic()) {
-            vlog(
-              clusterlog.warn,
-              "can't find {} in the metadata cache",
-              model::tx_manager_nt);
+            vlog(clusterlog.warn, "failed to create {}", model::tx_manager_nt);
             co_return find_coordinator_reply(
               std::nullopt, std::nullopt, errc::topic_not_exists);
         }
@@ -316,7 +351,8 @@ ss::future<bool> tx_registry_frontend::try_create_tx_topic() {
     return _controller->get_topics_frontend()
       .local()
       .autocreate_topics(
-        {std::move(topic)}, config::shard_local_cfg().create_topic_timeout_ms())
+        {std::move(topic)},
+        config::shard_local_cfg().create_topic_timeout_ms() * partitions_amount)
       .then([](std::vector<cluster::topic_result> res) {
           vassert(res.size() == 1, "expected exactly one result");
           if (res[0].ec == cluster::errc::topic_already_exists) {

diff --git a/src/v/cluster/tx_registry_frontend.h b/src/v/cluster/tx_registry_frontend.h
@@ -35,6 +35,8 @@ class tx_registry_frontend {
       ss::sharded<cluster::tx_coordinator_mapper>&,
       ss::sharded<features::feature_table>&);
 
+    ss::future<bool> ensure_tx_topic_exists();
+
     ss::future<find_coordinator_reply>
       find_coordinator(kafka::transactional_id, model::timeout_clock::duration);
 

diff --git a/src/v/kafka/server/server.cc b/src/v/kafka/server/server.cc
@@ -11,6 +11,7 @@
 
 #include "cluster/id_allocator_frontend.h"
 #include "cluster/topics_frontend.h"
+#include "cluster/tx_registry_frontend.h"
 #include "config/broker_authn_endpoint.h"
 #include "config/configuration.h"
 #include "config/node_config.h"
@@ -1566,6 +1567,16 @@ list_transactions_handler::handle(request_context ctx, ss::smp_service_group) {
         return true;
     };
 
+    auto& tx_registry = ctx.tx_registry_frontend();
+    if (!co_await tx_registry.ensure_tx_topic_exists()) {
+        vlog(
+          klog.error,
+          "Can not return list of transactions. Failed to create {}",
+          model::tx_manager_nt);
+        response.data.error_code = kafka::error_code::unknown_server_error;
+        co_return co_await ctx.respond(std::move(response));
+    }
+
     auto& tx_frontend = ctx.tx_gateway_frontend();
     auto txs = co_await tx_frontend.get_all_transactions();
     if (txs.has_value()) {

diff --git a/tests/rptest/tests/transaction_kafka_api_test.py b/tests/rptest/tests/transaction_kafka_api_test.py
@@ -30,7 +30,8 @@ def __init__(self, test_context):
                                  "tx_timeout_delay_ms": 10000000,
                                  "abort_timed_out_transactions_interval_ms":
                                  10000000,
-                                 'enable_leader_balancer': False
+                                 "enable_leader_balancer": False,
+                                 "transaction_coordinator_partitions": 4
                              })
 
         self.kafka_cli = KafkaCliTools(self.redpanda, "3.0.0")
@@ -136,6 +137,11 @@ def test_describe_transactions(self):
                 tpoic_partition = f"{topic}-{partition}"
                 assert tpoic_partition in expected_partitions
 
+    @cluster(num_nodes=3)
+    def test_empty_list_transactions(self):
+        txs_info = self.kafka_cli.list_transactions()
+        assert len(txs_info) == 0
+
     @cluster(num_nodes=3)
     def test_list_transactions(self):
         producer1 = ck.Producer({