1.6.0-beta8 November 18th, 2026
Akka.NET 1.6.0-beta8 adds TLS certificate rotation without a restart on both DotNetty and Artery, brings Akka.Reminders into the repository, and fixes a PersistAll completion bug plus two recurring-reminder ordering bugs. Read the v1.6 getting started guide for the full upgrade path.
Akka.IO / TLS
TlsCertificateSourcepicks the certificate a node presents from a folder, a certificate store by common name, or a callback, and picks up a renewed certificate without a restart. A folder or store source is re-read in the background at most once every 10 seconds and the result is reused in between; a callback is called every time. Startup waits for the first look-up, and a source with nothing usable throwsConfigurationExceptionso the node never binds. A connection that cannot get a certificate fails alone and the node stays up. (#8966, relates to #7663)- Certificate errors are logged at most once a minute per cause, so one bad certificate does not log a Warning for every connection. DotNetty and Artery now share
Akka.IO.TlsCertificateRotationfor that limiter, the startup check and the one-minute poll interval, but each transport keeps its own instance so one transport never hides another's errors. (#8982) - On Artery's provider path, a failed certificate look-up now surfaces as an internal
TlsCertificateUnavailableExceptionthat wraps the original cause, instead of writing a flag into the caller's own exception object. The only visible difference is one more--->level in the stack trace of Akka.Streams' "Restarting graph due to failure." warning. Fixed-certificate TLS and DotNetty are unaffected. (#8982)
Akka.Remote / DotNetty
- DotNetty TLS certificates now rotate without a restart.
DotNettySslSetup.FromSource, the HOCON keyscertificate.common-nameandcertificate.path(a folder),SslOptions.CertificateSourceandSslCertificateOptions.CommonNamewire aTlsCertificateSourceinto the transport and Akka.Hosting. (#8967) - Artery rotates TLS certificates through the same
TlsCertificateSource. Three new factories onArteryTlsSettings-Mutual(TlsCertificateSource, TlsPeerPolicy), a two-policyMutualoverload, andServerOnly(TlsCertificateSource, TlsPeerPolicy)- take a source, and the node asks it for a certificate each time it builds a connection. Akka.Remote.Hosting needs no new API:ArteryRemoteOptions.Tlstakes the profile. A startup look-up with nothing usable throwsConfigurationExceptionand the node never binds, and a connection that cannot get a certificate fails alone while the node stays up. (#8979, relates to #7663) - Behavior change:
SslOptions.CustomValidatornow runs when the certificate comes fromSslOptions.CertificateOptions(a filePathor aThumbprint). It used to run only withSslOptions.X509Certificate; with the other two it was silently ignored, so a security check you configured never ran. A validator that was ignored can now fail a TLS handshake, and on the dialing side a failed handshake shuts theActorSystemdown. WithEnableSsl = true, aCustomValidatorand no certificate in the options,WithRemotingnow throwsConfigurationException. The certificate also loads whenWithRemotingruns rather than when theActorSystemstarts, so a missing file or thumbprint fails earlier. See the v1.6 breaking-change ledger. (#8964)
Akka.Reminders (new module)
- Akka.Reminders now ships from this repository as
Akka.Remindersand its storage packages, replacing the standaloneAaron.Akka.Reminderspackages. It provides durable, re-entrant reminders for Akka.Cluster.Sharding with SQL Server, PostgreSQL, SQLite and in-memory storage, a cluster-singleton scheduler, and Akka.Hosting integration. Migration: the package ID changed, so update yourPackageReferenceand namespaces. (#8946) - Opt-in retry-until-acknowledged delivery: a reminder can require an explicit acknowledgement before it counts as delivered, with exponential backoff between attempts. The policy is persisted in all four storage backends, and existing signatures and default-policy wire layouts are unchanged. Opted-in schedules, responses and listings use versioned manifests, so upgrade every potential scheduler host and policy-message reader before enabling the feature - older schedulers must not take over while opted-in rows exist. Manual migration scripts for each provider live in
src/contrib/reminders/docs/migrations/retry-policy/, and every provider and potential scheduler host must be recreated after migrating because schema capabilities are cached. (#8973) - An early send of a recurring reminder now ends every older live occurrence atomically with the newer delivery, suppresses those rows for the rest of the fetch pass, and prevents a stale row from writing them back as Pending. Previously only the previous occurrence was ended, so a series with an interval below
MaxSlippagecould deliver stale occurrences after a newer one during a shard-region outage. (#8971, #8976, fixes #8965) - That cleanup is capped at 256 older occurrences per send, nearest first. A 1 ms interval would otherwise put thousands of expiries in one commit, and a sub-millisecond interval could put millions into the scheduler's memory. Below roughly a 20 ms interval at the default 5 s
MaxSlippage, an older occurrence more than 256 intervals back can still be delivered after a newer one during a shard-region outage. (#8975, #8976)
Akka
Unwatchnow clears the sole watcher from an optimized actor state.DefaultActorState.RemoveWatchedBynever cleared its optimized slot, so a target kept a reference to an actor that had unwatched it.FullActorStatealready removed watchers correctly. No public API or wire-format change. (#8936)
Akka.Persistence
- The
PersistAllcompletion callback no longer runs when the journal rejects the batch.OnPersistRejectedfired for each event and then the completion callback ran anyway, so a handler saw a completion for a batch that was never persisted. This affected all eightPersistAll/PersistAllAsyncoverloads. Journal write failures are unchanged: the actor stops before the completion's message is processed. (#8949, fixes #8948)
Akka.Remote / Artery
- Artery's inbound stage now reports a framing error when a peer closes after a partial connection preamble, frame length or frame body, instead of silently completing the connection. Empty connections and complete frame boundaries still close cleanly, and cancellation and upstream failure keep their existing semantics. Artery is new in v1.6, so this changes unreleased behavior only. (#8956)
Serialization
- Native module replacements can preserve legacy serializer type names through an optional type-alias map, so a ported serializer keeps reading data written under the original assembly-qualified name. Each declared name resolves to an existing module row using the declared type's assembly simple name and the existing strict name-matching rules; subclasses are not matched. Missing or ambiguous aliases, non-serializer types and conflicting registrations are rejected. No serializer rows, ids or configuration switches are added. (#8920)
Akka.Hosting
Akka.Hosting,Akka.Cluster.Hosting,Akka.Remote.Hosting,Akka.Persistence.HostingandAkka.Persistence.Embedded.Hostingnow ship their own module READMEs on nuget.org instead of the generic Akka.NET one. Relative links were rewritten to absolute GitHub permalinks because nuget.org cannot resolve relative paths. (#8945)
Docs
- New TLS certificate rotation article covers folder, store and callback sources, when a renewed certificate is used, what happens when a look-up finds nothing usable, and the error-logging behavior. Both the DotNetty transport and Artery rotate certificates the same way. (#8968, relates to #7663)
Internal
MultiNodeSpecnow terminates a replacement actor system when initialization fails, including on caller cancellation, and preserves the original exception. (#8955)RemotePingPongvalidates delivery and compares authenticated TCP, and measures latency at a scheduled offered load with per-run reporting that separates generator lateness, dispatch delay, send-to-echo and scheduled-to-echo latency. (#8953, #8957)- Artery's DeathWatch ACK-loss test now witnesses TCP system-message retries by counting dropped ACKs per peer UID and cumulative sequence. (#8954)
ShardedDaemonProcessSpecswaits for sharded daemon restarts with one overall deadline instead of back-to-back 3-second waits, fixing a Windows CI timeout. (#8981)- Slopwatch upgraded to 0.5.0 with a regenerated schema-v2 baseline. (#8980)
- The Serialization V2 port-comparison test kit now compares the root message's members in its structural fallback and checks the exact runtime type before the equality fast path, and supports shared system names. Test-only; no production API, bindings or wire bytes change. (#8916)
1 contributor since release 1.6.0-beta7
| COMMITS | LOC+ | LOC- | AUTHOR |
|---|---|---|---|
| 22 | 44500 | 1088 | Aaron Stannard |