3.36.2

Added

• Configurable Redis connection buffer sizes via read-buffer-size and write-buffer-size on any Redis configuration (for example as.redis.read-buffer-size and as.redis.write-buffer-size). Both default to 0, which uses the client default of 32 KiB per buffer. Deployments with a large number of connections (such as the Application Server’s per-subscription Pub/Sub connections) can lower these to reduce memory usage.
• MONITORING_STOPPED value for the alert notification status. For ALERT_GATEWAY_DISCONNECTED alerts, the Alert Routing Server now verifies the gateway’s live connection state with the Gateway Server when an alert resolves, and emits MONITORING_STOPPED (instead of RESOLVED) when monitoring stopped while the gateway was still offline, so receivers no longer read it as “gateway back online”.
• dcs.gcls.semtech-rjs.lns-uri and dcs.gcls.semtech-rjs.lns-trust configuration options. When set, gateways claimed through the Semtech RJS are configured with this LNS directly, instead of being redirected to the CUPS. This fixes claiming of The Things Indoor Gateways, which cannot verify the cross-signed Let’s Encrypt certificate chain served by the CUPS.
• events.redis.store.event-type-names configuration option: an allowlist of event names written into the per-event-type streams that group subscribers (such as the Network Operations Center) consume. Empty by default, which writes all events and preserves existing behavior. Setting it to the consumed set stops writing event-type streams that nothing reads, reducing XADD load on the events Redis. In deployments where events are published from a different process than the group subscriber, the publisher processes must be configured with the names their subscribers consume.

Changed

• Alert notification webhook, email and SMS receivers may now receive the MONITORING_STOPPED status for gateway disconnection alerts. Webhook consumers that switch exhaustively on the status should handle the new value.
• The events Redis store now refreshes each per-entity event stream’s TTL at most once per half of the configured entity history TTL, instead of on every published event, to reduce PEXPIRE command load (and therefore CPU) on the events Redis. Event history retention is unchanged.
• The rate limit for reading an end device (the Get RPC on the Identity Server, Network Server, Application Server and Join Server end device registries) is now applied per application instead of per end device. A client that reads many devices in the same application now shares a single rate limit for those reads, rather than getting a separate limit per device.

Fixed

• Identity Server user and organization reads no longer scan the entire accounts table: lookups against the user_accounts/organization_accounts views are now scoped by the account’s tenant, so the accounts (tenant_id, uid) index is used.
• Applying the MAC settings profile values to the device’s MAC state during the join procedure (OTAA) or factory reset (ABP).
• Federated (OIDC/SSO) login no longer fails with linked_user (linked user not found) after a linked local user is purged. Purging a user now also removes its external-user associations, so no orphaned external_users row is left to block a later SSO login with the same identity.
• The as.webhook.fail event now includes the top-level correlation IDs of the upstream message when webhook retries are enabled. Previously the correlation IDs were lost when the message passed through the webhook queue, which prevented the FindRelatedEvents API from tracing the chain of events related to the failed webhook delivery.