nano v2.0.7
nano v2.0.7 brings full high availability to multi-node installs: several app servers, automatic PostgreSQL and Valkey failover, and rolling upgrades.
v2.0.7 includes everything in v2.0.6.
To upgrade, follow Upgrade a self-hosted deployment.
Security
This release makes JWT revocation fail closed in high-availability mode, requires publish permission to create a playbook that is already live, and restricts who can change a notebook's visibility. We don't publish exploit details. To report a vulnerability, see Support.
Highlights
A multi-node install can now lose any one server and recover on its own, with at most a short pause for some operations (below). See High availability.
- Several app servers. Run any number of stateless app and search servers behind your load balancer. Each answers
/readyz, which reports 503 while the server is draining or broken, so the load balancer only sends traffic to servers that can serve it. See Add an app server. - PostgreSQL failover. With three PostgreSQL servers, PostgreSQL fails over automatically (Patroni and etcd). Replication is synchronous by default, so a failover keeps every acknowledged write. When the primary is lost, writes, sign-ins and searches fail while a standby is promoted: about 20–25 seconds when the primary's server dies, a few seconds for a planned switchover. The UI and views that are already loaded stay up, and ingest over HEC and HTTP is unaffected. No acknowledged write or event was lost in our tests. See PostgreSQL servers.
- Valkey with Sentinel. When the Valkey primary is lost, a replica took over in about 7 seconds in our tests. Sessions that were logged out stay logged out throughout.
- ClickHouse replica failover. Each server reaches ClickHouse through a local HAProxy that checks every replica actively, so losing a replica costs at most a couple of seconds of failed searches. Ingest and alerting continue, and writes are never replayed on a retry.
- Scheduled backups fail over. Every app server runs the backup scheduler, and exactly one takes each scheduled backup. If that server is lost, another one takes over.
- Ingest stays in rotation during API outages. An ingest server keeps answering its load-balancer health check while the API or PostgreSQL is briefly unavailable, so a short control-plane outage no longer takes every ingest server out of rotation.
- Load-balancer examples. The bundle includes tested HAProxy and keepalived examples in
deploy/compose/examples/load-balancer/. With keepalived, the floating address moves to the second load balancer in about 3 seconds. - Rolling upgrades.
nano upgradeupgrades one server at a time. It drains each app server from the load balancer and waits for it to rejoin before moving on. PostgreSQL replicas are upgraded first, then nano switches over and upgrades the old primary. See Upgrade. - Plan a deployment with your AI agent. The
nano-deployskill plans a full-HA layout on VMs you already have, or picks VMs on AWS, Azure or GCP, then walks your agent through the install. See Deploy with your AI agent.
Fixes
- Pull connectors and scheduled backups now run on multi-node installs. Before, they never started there.
- JWT revocation fails closed in high-availability mode. If the shared cache can't be reached, the API answers 503 with
Retry-Afterand never accepts a revoked token. - The jobs leader and the alert pipeline keep running across brief database outages.
- Creating a playbook with status
liverequiresplaybooks:publish, the same as publishing an existing one. - Notebook visibility. Unlinking a notebook from a case makes it private to its owner again, or shared if it still has explicit shares. Only the owner can change a standalone notebook's visibility.
Upgrade notes
- Single-node and single-app installs are unchanged. The upgrade renders the same services as before.
- Converting to several app servers or HA PostgreSQL is a topology change that you acknowledge. The installer first refuses and prints the current identity; rerun it with
--accept-topology-migration <identity>. No data moves. Follow Add an app server and PostgreSQL servers. - Outbound access. Multi-node installs pull third-party images (HAProxy, etcd, Valkey and others) by digest from Docker Hub, quay.io and ghcr.io, as well as nano's own registry. Allow outbound HTTPS to them from every server. The preflight check stops early if one is unreachable.
- What a lost server still affects. Events buffered on a lost ingest server wait until it returns. Losing both replicas of a ClickHouse shard, or most Keeper servers, stops writes. Your load balancer must itself be redundant; see Load balancer options.
- Migrations are forward-only and run before the new services start.
- Upgrade with
scripts/nano upgrade, as described in Upgrade a self-hosted deployment.
nano v2.0.8
nano v2.0.8 includes security fixes, an enforced ingestion pause, a more reliable CrowdStrike FDR source, and install and high-availability fixes.
nano v2.0.6
nano v2.0.6: security fixes, a CrowdStrike FDR pull source, scheduled automatic backups and an ingest durability fix. All Self-run installs should upgrade.