nano v2.0.11
nano v2.0.11: a Helm security fix, Okta single sign-on, PostgreSQL 16 support, and install, upgrade and HA fixes for Docker Compose and Kubernetes.
v2.0.11 includes everything in v2.0.10.
To upgrade, follow Upgrade a self-hosted deployment.
Known issue (fixed in v2.0.12)
On Self-run Docker Compose installs, scripts/nano upgrade from v2.0.10 to v2.0.11 fails its preflight with two false positives:
- A registry-login check for the public Vector image on
ghcr.io. - An OCSF migration-ledger check, on installs with OCSF.
Nothing is wrong with the install. Fixed in v2.0.12: upgrade from v2.0.10 straight to v2.0.12 with scripts/nano upgrade. To upgrade to v2.0.11 itself, rerun the v2.0.11 installer command on each host instead; on a single-host install, rerun its one install command. Your data is preserved.
Helm installs are not affected.
Security
- Helm: Vector's in-cluster OTLP port. Before v2.0.11, Vector's OTLP and native listeners were reachable inside the cluster without a credential, and only a NetworkPolicy kept other pods out. Many clusters don't enforce NetworkPolicy by default (GKE Standard, default EKS and AKS), so any pod could send events. From v2.0.11 these listeners accept connections only from inside the Vector pod, behind a guard that requires a per-install token sent by the ingest edge. See Kubernetes: Network policy.
- Single sign-on hardening. A verified email now requires a real
emailclaim. A verified domain never overrides an identity provider that says the email is not verified, and free-mail and public-suffix domains can't be marked verified.
Fixes and improvements
- Okta single sign-on. nano now reads
emailandemail_verifiedfrom the provider's userinfo endpoint when the ID token leaves them out, which Okta does. Admins can set verified email domains for Okta and other OIDC providers. Refused sign-ins show the reason in the audit log. See Single sign-on. - PostgreSQL 16. nano now runs on PostgreSQL 16 and later. Before migrating, it checks the server version and the extensions it needs (
pg_trgm,pgcrypto,uuid-ossp). On Azure Database for PostgreSQL, allow those extensions first. See Kubernetes: Managed PostgreSQL. - Helm.
- IP and geo enrichment now load on high-availability ClickHouse.
- A slow first migration no longer makes the install fail.
helm upgrade --reuse-valuesfrom v2.0.10 works.- New guides for GKE and AKS, and options to keep the AWS load balancers internal.
- Docker Compose.
- A data disk that misses its mount after a crash now mounts itself and then starts Docker.
- The cluster health check fails when Docker or the data disk is down.
- The PostgreSQL high-availability identity is recorded on the first PostgreSQL server.
- The generated firewall no longer opens port 443 to everyone by default.
- Getting help. New
nano support-bundlecommand, and Report a bug and Get support in the Help menu. See Getting help. - Search. Filtering on OCSF unmapped fields from the fields panel works again.
Known issues
These are fixed in v2.0.12. Upgrade to v2.0.12 or later.
- Upgrading from v2.0.10 with
scripts/nano upgradefails its dry run on Docker Compose (Self-run) withregistry-login ... not logged in to ghcr.ioand, on OCSF installs,ledger-clickhouse ... not in target ocsf/193,ocsf/194. Both are false positives and nothing is changed. Upgrade to v2.0.11 by running each server's install command again instead, as in High availability: From v2.0.9 or earlier (on one server, run your install command again), or upgrade to v2.0.12 withscripts/nano upgrade. - A fresh multi-server install can stop at the first app server's database migration. The installer waits at
clickhouse-migratewith no progress, and the migration log ends at "Applying open-init snapshot". If nothing changes for 5 minutes, stop the install (Ctrl-C; nothing is half-applied), check that PostgreSQL has a leader and a streaming synchronous standby (patronictl liston apostgresserver), and run the same install command on that server again. v2.0.12 ends a stuck migration on its own and reports why. - The data-disk self-heal timer isn't installed by the installer. Install it on each server as in Prepare the data disk. v2.0.12 installs it.
- The message after the PostgreSQL servers' install says
sudo ./install.sh --record-pg-ha-identity. Run it withoutsudo, as the user that installed nano:cd ~/nano && ./install.sh --record-pg-ha-identity. scripts/nano support-bundlereports no install ID, and on several servers collects only the server it runs on unless that server can reach the others over SSH. See Getting help.- OTLP during an app server's upgrade. With several app servers, an OTLP request could get a
500or stall while one app server restarted; HEC and HTTP were not affected. Make OTLP senders retry. v2.0.12 retries the check that OTLP needs on another app server. - Helm on Azure flexible server: the first migration fails with
permission denied for schema public. Schemapublicbelongs toazure_pg_adminthere, and nano's checks before the migration didn't test it. As the admin user, connected to thenanosiemdatabase, runALTER SCHEMA public OWNER TO nanosiem;, then run the samehelm upgrade --installagain. See Managed PostgreSQL requirements. v2.0.12 checks it first and names the fix. - Helm: OTLP writes get a
500during a PostgreSQL failover, for as long as the api pods are unready (about 48 seconds on Azure). OpenTelemetry exporters drop a500. HEC and HTTP are not affected. Buffer OTLP in a collector that retries on all5xx. v2.0.12 answers503withRetry-Afterinstead, which exporters retry. - Helm: the nano-deploy skill's health check stays at
UNHEALTHY: chafter a ClickHouse pod restart, withdistributed DDL entries pending > 10 min, although nothing is wrong. See What to expect during a failure. v2.0.12's check only warns about such an entry. - Helm: the ClickHouse operator image is pulled by tag, not digest like nano's own images. v2.0.12 pins it by digest.
Upgrade notes
- Helm: Vector native port. The Vector native port is now off by default. If you forward events from another Vector in the cluster to port 6000, set
vector.native.enabled: true. - Helm: a short OTLP gap during the upgrade. While the ingest edge and Vector roll out from v2.0.10, some OTLP requests can get a
401, and OpenTelemetry exporters drop those batches. HTTP and HEC are not affected. Pause OTLP senders during the upgrade to avoid any loss. - Helm: existing secrets. If you manage the chart's secret yourself (
secrets.existingSecret), addvector_otlp_hop_token(32 or more characters fromA-Z a-z 0-9 _ -) before upgrading. - Docker Compose: firewall script. If you generate the host firewall with the planner and you have HAProxy hosts, pass
--edge-source-cidr. Remove any oldallow 443/tcprule first.
nano v2.0.12
nano v2.0.12 adds optional Self-run monitoring, fixes the Compose upgrade tool, checks managed PostgreSQL permissions first and fixes Helm and OTLP.
nano v2.0.10
nano v2.0.10 is the minimum licensed Self-run release. It adds a per-install ID to license check-ins, upgrades from the signed bundle and fixes HEC limits.