batchlog_manager: set timeout on writes

Set a timeout on writes of replayed batches by the batchlog manager.

We want to avoid having infinite timeout for the writes in case it gets
stuck for some unexpected reason.

The timeout is set to be high enough to allow any reasonable write to
complete.
This commit is contained in:
Michael Litvak
2025-06-25 14:10:52 +03:00
parent 7150632cf2
commit 74a3fa9671
2 changed files with 3 additions and 1 deletions

View File

@@ -244,7 +244,8 @@ future<> db::batchlog_manager::replay_all_failed_batches(post_replay_cleanup cle
// send to partially or wholly fail in actually sending stuff. Since we don't
// have hints (yet), send with CL=ALL, and hope we can re-do this soon.
// See below, we use retry on write failure.
return _qp.proxy().send_batchlog_replay_to_all_replicas(std::move(mutations), db::no_timeout);
auto timeout = db::timeout_clock::now() + write_timeout;
return _qp.proxy().send_batchlog_replay_to_all_replicas(std::move(mutations), timeout);
});
}).then_wrapped([this, id](future<> batch_result) {
try {

View File

@@ -45,6 +45,7 @@ public:
private:
static constexpr uint32_t replay_interval = 60 * 1000; // milliseconds
static constexpr uint32_t page_size = 128; // same as HHOM, for now, w/out using any heuristics. TODO: set based on avg batch size.
static constexpr std::chrono::seconds write_timeout = std::chrono::seconds(300);
using clock_type = lowres_clock;