Redpanda Release Notes
This page lists the changes in each Redpanda release from version 26.1.18 onward, organized by version. For a curated summary of the major features in this release line, see What’s New.
v26.1.20 (2026-11-04)
Features
- Admin API
-
-
The
/v1/cluster_config/schemaendpoint now includes each property’s default value, numeric bounds where applicable, and enterprise license restriction details. -
sasl_mechanisms,log_cleanup_policy,transaction_coordinator_cleanup_policy,enable_consumer_group_metrics, andhttp_authenticationnow report their accepted values in the/v1/cluster_config/schemaresponse, and, forsasl_mechanisms, which values require an enterprise license.
-
Bug fixes
- Cloud Storage
-
-
Cloud Topics L0 garbage collection on Azure Blob Storage no longer ignores
cloud_topics_short_term_gc_minimum_object_age. No data was lost, because garbage collection deletes only objects whose data is already in L1.
-
- Cloud Topics
-
-
Cloud Topics partitions no longer answer OffsetForLeaderEpoch with an undefined offset for a leader epoch that is older than the local log and unknown to the L1 metastore. The start of the local log is returned instead, matching the other storage modes and Kafka.
-
The Cloud Topics L0 write scheduler no longer has a race in which concurrent upload group splits and merges could corrupt the group layout and degrade upload throughput scaling.
-
The Cloud Topics metastore topic now reaches
internal_topic_replication_factoras soon as enough brokers have joined a new cluster, instead of staying at a single replica for up to 10 minutes.
-
- Cluster
-
-
Brokers on ARM (aarch64) no longer crash in rare cases when the CPU profiler or stall detector samples a thread mid stack switch, for example while collecting a debug bundle.
-
A broker no longer hits an out-of-memory abort by forwarding controller join and snapshot requests to itself after it steps down as controller leader.
-
A concurrent cross-shard move during a topic deletion no longer triggers a
vassert()in theshard_placement_table. -
A broker killed mid-write of the configuration cache no longer leaves a torn file that aborts future boots of
redpandauntil the file is removed. -
The
health_managerno longer stops ticking after acontrollerleadership transfer. -
Changes to
replication.factorno longer silently fail to apply to some partitions in a topic, including whenhealth_monitorapplies them to internal topics. -
Deleting a very large topic, with more than 1,000 partitions per CPU core, no longer makes a broker unresponsive until it restarts. This issue was rare.
-
Brokers no longer report the old leader of a partition after its topic is deleted and created again. Previously, some Kafka clients then stopped producing to that partition.
-
- HTTP Proxy
-
-
Steady HTTP Proxy produce traffic no longer keeps pushing back
produce_batch_delay_ms, which delayed requests by up to several seconds.
-
- Iceberg
-
-
Iceberg commits now count
status=deletedmanifest entries correctly when they create a new manifest list as part of a new Iceberg snapshot. Previously, the catalog could reject commits when another writer to the table, like a compaction process, deleted data files and wrote manifests withstatus=deletedentries. -
Unity Catalog no longer refuses Iceberg commits after an external engine compacts or deletes data in the table.
-
Iceberg translation now supports a UUID logical type backed by a
fixed(16)physical type in Avro schemas.
-
- Kafka API
-
-
A consumer group member that re-sends JoinGroup while its previous JoinGroup is still pending, for example after a dropped connection during a rebalance, no longer has its connection closed with "updating non-joining member". The new request takes over and the earlier one is answered with
REBALANCE_IN_PROGRESS, matching Apache Kafka and ending the client reconnect loop. -
A Kafka request with an oversized length field no longer aborts the broker before the request is validated.
-
Transaction abort handling (
try_abort) no longer hits state-lock deadlocks that could leave a broker hung. -
An
IncrementalAlterConfigsrequest with aDELETEoperation that includes a value is no longer rejected.redpandanow ignores the value. -
Transactional offset commits no longer fail with an unknown server error when they race a
__consumer_offsetsleadership change. They now returnNOT_COORDINATOR, and clients retry them.
-
- Schema Registry
-
-
Schema Registry no longer returns HTTP 500 on its first request after a broker restart while controller leadership is still being established.
-
Negative Protobuf message indexes no longer crash the broker during schema ID validation or Iceberg translation.
-
Schema Registry now rejects base64-serialized Protobuf schemas with out-of-range dependency indexes instead of crashing the broker.
-
- Shadow Link
-
-
The shadow link Consumer Group Shadowing task no longer logs an error and briefly reports a faulted state when a source cluster broker leaves the cluster metadata.
-
Schema Registry no longer ignores
CREATEauthorization failures for_schemaswhen a shadow link is active. If_schemasdoes not exist yet, a configured SCRAM user now needsCREATEpermission to create it. -
While a shadow link is active, creating topics now requires superuser only for topics that the link mirrors or would auto-create. Other topics follow normal ACLs.
-
- Storage
-
-
Broker restarts after a crash are no longer slow. Previously, on every later restart, the broker re-read the segment that each partition was writing at the time of the crash, and logged
Recovered:for it. The first restart after upgrading fixes these segments.
-
- Tiered Storage
-
-
Changing
cloud_storage_segment_max_upload_interval_secnow reports that a broker restart is required. Previously, the change was accepted with no restart prompt, but brokers kept uploading on the old interval until they were restarted. -
The adjacent segment merger no longer asks
rm_stmfor aborted transactions past the offsets it has applied. -
Tiered Storage cache accounting no longer underflows the cache size counter when a trim races a concurrent put, or when a file was already removed. Previously, the cache then believed it was full, blocked new entries, and evicted everything it could. The cache also no longer grows past its configured limit because of a staged file that was never counted.
-
Tiered Storage partitions now return the cloud start offset rather than the local start offset when a replica’s local log extends below the cloud start offset. Previously, the local start offset caused problems mainly with follower fetching.
-
Improvements
- rpk
-
-
The
rpk shadow deletehelp now describes--forcecorrectly: it removes the shadow link even when shadow topics are active, and those topics become regular, writable topics that keep their data, with no further replication.
-
- Storage
-
-
Local disk space management now uses less CPU on brokers with a large number of segments.
-
v26.1.18 (2026-09-24)
Bug fixes
- Cloud Storage
-
-
The internal HTTP client no longer performs an out-of-bounds read when decoding a URI that ends with a truncated percent-escape.
-
- Cloud Topics
-
-
The write-request scheduler no longer consumes a full CPU core on shards whose backlog cannot be uploaded.
-
- Iceberg
-
-
Iceberg producer throttling now uses
max_kafka_throttle_delay_msas the throttle ceiling andquota_manager_gc_secas the producer GC window. These two settings were previously flipped at the point of use.
-
- Kafka API
-
-
A broker connection in the internal Kafka client (used by Schema Registry, HTTP Proxy, and Shadow Link) no longer gets stuck unable to complete API version negotiation. Previously such a connection could fail repeatedly with
broker_not_availableuntil the socket dropped for an unrelated reason. -
A response that arrives after its request has already timed out no longer desyncs the internal Kafka client connection’s protocol framing. Previously this could cause cascading disconnects and stalled requests on that connection.
-
The Kafka quota manager no longer triggers a use-after-free during shutdown in its garbage collection.
-
- Schema Registry
-
-
JSON Schema compatibility checks now handle schemas that use fragment-only or fragment-bearing
$idvalues. Previously these checks could resolve$refvalues against the wrong subschema. Compatibility verdicts for affected schemas may change. -
JSON Schema compatibility checks no longer crash on schemas whose
$reftargets a boolean or a non-schema value. -
A JSON Schema compatibility check no longer aborts the broker when a required property has a boolean subschema.
-
Produce requests are now rejected if any schema-ID-validated key or value in a batch is invalid. Previously a batch could be accepted when only its last validated component was valid, which persisted invalid records.
-
- Security
-
-
Describing or deleting ACLs on clusters with many distinct ACL resource patterns no longer causes oversized memory allocations.
-
OpenSSL is upgraded to 3.5.8, which addresses reported security vulnerabilities.
-
c-ares is upgraded to 1.34.8, which addresses CVE-2026-33630.
-
- Storage
-
-
Records are no longer silently dropped from compacted topics after a partly successful log append.
-
- Tiered Storage
-
-
Metadata spillover no longer gets permanently stuck on partitions with small, frequent segments. Previously it could repeatedly log
Can’t apply spillover_cmdand leave orphaned spillover manifest objects in the bucket.
-
Improvements
- Cloud Storage
-
-
The internal HTTP client now fails a request with a timeout when the connection makes no I/O progress for 120 seconds. Previously the request stalled until TCP retransmission gave up, about 15 minutes, when a remote endpoint hung.
-
- Cluster
-
-
Reconnection logic is hardened when inter-broker heartbeat requests hit a connection timeout.
-
Cross-shard partition movements now transfer the local
log_eviction_stmsnapshot. Previously the missing snapshot led to full local log replays.
-
- Schema Registry
-
-
A new
redpanda_schema_registry_loaded_offsetmetric surfaces the last offset from the Schema Registry’s internal_schemastopic applied to the local store.
-
Release notes for older versions
Release notes for versions before 26.1.18 are published on GitHub. See the Redpanda releases page.