Initial commit

Signed-off-by: teodordelibasic-db <teodor.delibasic@databricks.com>

Author: teodordelibasic-db

rust/NEXT_CHANGELOG.md

@@ -8,6 +8,7 @@
 
 ### Bug Fixes
 
+- **Arrow Flight — acknowledgment deadlines remain active during graceful close** (Beta): a server close signal no longer pauses or extends an already-running batch deadline. Repeated close signals preserve the earliest recovery deadline, while a valid ACK already ready at either deadline is applied before recovery is considered.
 - **Arrow Flight — acknowledgment deadlines now start when work becomes pending** (Beta): no timer runs while the stream is idle. The timeout is an absolute deadline for the oldest pending batch: acknowledgments that leave that batch pending do not refresh it, replayed batches receive a fresh deadline on their recovered connection, and malformed or non-progressing responses cannot indefinitely postpone recovery. A valid acknowledgment already ready at expiry is applied before recovery is considered.
 - **Arrow Flight — invalid acknowledgment watermarks are rejected** (Beta): ACK progress is now monotonic, so delayed or duplicate responses cannot move the durable watermark backward. A response claiming more records than were actually submitted on the active connection is rejected without making buffered, unsent records appear durable.
 ### Documentation

rust/README.md

@@ -804,8 +804,9 @@ For Arrow Flight streams *(Beta)*, `server_lack_of_ack_timeout_ms` is an
 absolute limit on how long a batch may remain pending, not an inactivity
 timeout. No timer runs while the stream is idle. Each batch's timer starts when
 it becomes pending and is not refreshed when earlier batches are acknowledged.
-Partial acknowledgments do not extend it. If recovery replays pending batches
-on a new connection, their deadlines restart relative to that connection.
+Partial acknowledgments and graceful close do not pause or extend it. If
+recovery replays pending batches on a new connection, their deadlines restart
+relative to that connection.
 Configure the timeout together with `max_inflight_batches` so the server can
 acknowledge a full allowed backlog within the timeout.
 

rust/sdk/src/arrow_configuration.rs

@@ -70,8 +70,9 @@ pub struct ArrowStreamConfigurationOptions {
     ///
     /// No timer runs while there are no pending batches. Each batch's timer starts when it
     /// becomes pending and is not refreshed when earlier batches are acknowledged. Partial
-    /// acknowledgments do not extend it. Configure this timeout and `max_inflight_batches`
-    /// so the server can acknowledge a full allowed backlog within the timeout.
+    /// acknowledgments do not extend it, and graceful close does not pause it. Configure
+    /// this timeout and `max_inflight_batches` so the server can acknowledge a full allowed
+    /// backlog within the timeout.
     /// After recovery, replayed batches receive a fresh deadline relative to the recovered
     /// connection. If a batch remains pending when its deadline expires, the stream is
     /// considered failed and recovery is triggered (if enabled).