nano v2.0.28
nano v2.0.28: the interactive install.sh fix, one ClickHouse user per workload, ingest no longer refused at the hourly query quota, and upgrade and Helm fixes.
v2.0.28 includes everything in v2.0.27. It is not a security release.
To upgrade, follow Upgrade a self-hosted deployment. v2.0.20 is a required stop: installs on a release older than v2.0.20 upgrade to v2.0.20 first. Read the Upgrade notes below first: the upgrade adds ClickHouse users, and installs on several hosts and Helm installs with their own Secret have a step to do before they upgrade.
Fixes
- An interactive
install.shno longer crashes at the storage tiering question. In v2.0.27, answering "Enable cold storage tiering?" in a terminal failed withio.UnsupportedOperation: File or stream is not seekable. The prompt now works in a terminal. Unattended installs were never affected. - Ingest is no longer refused on busy installs. Every nano ClickHouse user had the same hourly cap of 100,000 queries. Busy installs reached it and ingest failed with
QUOTA_EXCEEDED. nano's own service users (the app, Vector, direct OCSF writers and migrations) no longer have an hourly query cap. Their use is still tracked. Raw SQL users keep their cap. - Agent Lane works on several hosts with private TLS (Self-run). Turning Agent Lane on and creating tokens returned
503 clickhouse_users_unverifiedon that topology, the default for several hosts. Re-runinstall.shon every app host after the upgrade. - The upgrade's ClickHouse lockout check flags only outside clients. On one host,
scripts/nano upgradeno longer stops for ClickHouse's own internal queries (0.0.0.0) or for nano's containers from before the network pin. It lists only clients the upgrade would newly lock out, so you no longer need--ack-clickhouse-network-lockoutfor nano's own addresses. - An edited Tenzir pipeline no longer blocks
install.shupdates. On git installs from get.nano.rs, an editedtenzir/pipeline.tqlor apipeline.tql.bakbackup stopped the update, andinstall.shreinstalled the old release and reported success. Now your edited pipeline is kept and stays live. Any other local change that blocks the update fails the install, lists the files and installs nothing.NANO_SKIP_GIT_UPDATE=1still reinstalls the checkout as it is. - The Helm egress NetworkPolicy no longer breaks DNS on clusters with a node-local DNS cache. In v2.0.27 it admitted DNS only to kube-dns pods, which cut DNS on stock GKE. Now it works on every cluster DNS path: NodeLocal DNSCache on GKE, LocalDNS on AKS, and CoreDNS on EKS and AKS. The new
networkPolicy.egress.dns.modedefaults toany, which allows DNS ports to any destination and opens no other port. See Egress NetworkPolicy. - No error on the signed-out login page. A signed-out page load no longer sends a session refresh that logged a
400error in the browser console.
Changes
- One ClickHouse user per workload. Each workload now connects to ClickHouse as its own user, with limits sized for it:
- Search runs as the new
nanosiem_searchuser. That covers the Search page, dashboard panels, field values, prevalence pages, reports, pivt, the MCP server and nano Desktop, plus AI searches. It has its own hourly quota, set at twice what the app can run at once for a whole hour, as a backstop against a runaway. Searches that reach it fail withQUOTA_EXCEEDEDuntil the hour rolls over. - Dictionary sources (prevalence, IP geo, IOC, custom enrichment, user registry) read as the new
nanosiem_dictuser, which can read only the tables they use. - Direct OCSF writers such as Tenzir have bounded reads: 300 seconds and 100 million rows per query.
- Backups and the Agent Lane query audit have their own small quotas.
- Search runs as the new
- SIEM Health reports ClickHouse quota use. It raises a high finding when a nano ClickHouse user had queries refused with
QUOTA_EXCEEDEDin the last hour, and a medium warning when one is past 80% of a capped hourly limit. - ClickHouse cluster secret on several hosts. The upgrade stages a shared cluster secret on every ClickHouse host. Nothing changes until you switch it on, and switching is single-shard only for now. With it on, a query that crosses shards runs as the user who started it, with that user's grants, row policies and quota, instead of as
nanosiem. See Upgrade notes. - Faster drilldown and table paging in Search. Clicking a value in a
top, timechart, flow or table view now requests only the columns the table shows, and loads field stats in the background. In our tests the first rows of a drilldown appeared in about 0.2 seconds instead of about 1.1 seconds. Table paging requests the table view too. - Clearer Agent Lane settings page (Self-run). Settings, Agent Lane has a short plain description and rows that no longer overlap. When the IP allow-list is missing or invalid, a setup card shows the config and commands for one host, several hosts and Helm. Skill install commands have copy buttons.
- The natural-language hint above the search box is gone.
Upgrade notes
- Required stop at v2.0.20. See above.
- New ClickHouse users and passwords. The upgrade adds
nanosiem_searchandnanosiem_dictand generates their passwords,CLICKHOUSE_SEARCH_PASSWORDandCLICKHOUSE_DICT_PASSWORD. ClickHouse gets only their SHA-256 hashes.-
One host:
scripts/nano upgradeandinstall.shgenerate both passwords in.envand set up the users. Nothing to do. -
Several hosts: every ClickHouse server needs both passwords and the new users before the app host switches over. Do these steps in order:
- On every host, verify the v2.0.28 installer package and extract it over the install directory as the user that owns it, not as root. This step matters: the upgrade runs the upgrader from the extracted release, and only the v2.0.28 upgrader knows about the new users. If you skip it, the v2.0.27 upgrader runs, checks nothing and sets up no new users; nano keeps running as before, but without the new users and limits.
- On the app host, run the new installer as
NANO_PREPARE_ONLY=1 ./install.sh --inventory <inventory> --role <its roles>. It generates the new keys in.envand starts nothing. - Copy that
.envto every host. - From the app host, run
scripts/nano upgrade --inventory …as usual. It upgrades the Keeper and ClickHouse hosts first, and refuses to start until every host's.envhas the new keys.
If you upgrade the hosts one by one with
install.shinstead, do steps 1 to 3, then run every Keeper and ClickHouse host before any app host. -
Helm: the chart's Secret generates both passwords. With
secrets.existingSecret, first addclickhouse_search_passwordandclickhouse_dict_password(32 or more random characters, no$), and setsecrets.workloadPasswordHashes.searchandsecrets.workloadPasswordHashes.dictto their SHA-256. Without them the render stops and says so. New api, search and jobs pods wait up to 10 minutes for the search user to log in on every ClickHouse shard, while the old pods keep serving. -
Dictionaries switch to
nanosiem_dictonce that login works on every ClickHouse server. Until then they read as before. On Helm the switch happens on the nexthelm upgrade. -
Helm rollback: once the dictionaries use
nanosiem_dict,helm rollbackto 2.0.27 removes that user and the dictionaries stop loading until 2.0.28 is applied again. Roll forward instead, or restore the pre-upgrade backup.
-
- Argo CD and other GitOps tools: use
secrets.existingSecret. With the chart-managed Secret, every render without cluster access, such as Argo CD's, generates new passwords and hashes. - Users sign in once after the upgrade. A browser session from before the upgrade lands on the login page once, on its first reload.
- Cluster secret (several hosts). The upgrade puts
NANO_CLICKHOUSE_CLUSTER_SECRETin the shared.envand stages it on every ClickHouse host. Cross-shard queries still log in asnanosiem, as before. On a single-shard cluster you can switch it on once every ClickHouse host runs v2.0.28: setNANO_CLICKHOUSE_CLUSTER_AUTH=secretin the app host's.env, copy it to every host, and re-runinstall.sh --role clickhouseon the ClickHouse hosts back to back, in a quiet period. Until the last one is re-run, queries between switched and unswitched hosts fail. The installer refusessecreton a cluster with more than one shard. To roll back a host in secret mode to an earlier release, setNANO_CLICKHOUSE_CLUSTER_AUTH=stagedand re-run the ClickHouse hosts first. - Egress NetworkPolicy
dns.mode. An existing values set withoutdns.moderenders asany. If you setnetworkPolicy.egress.dns.toon 2.0.27, addnetworkPolicy.egress.dns.mode: peersto keep your list.peerswith an empty list is refused. To turn the policy on, follow the steps in Egress NetworkPolicy. - Helm undersized-disk check. The refusal message now says it applies to a first install only, and names
sizing.allowUndersizedDisk. See Upgrades. - Extract the installer package as the install directory's owner. On registry installs, verify the package, then extract it over the install directory as the user that owns it, not as root.
scripts/nano upgraderefuses to run as root. See Upgrade a self-hosted deployment. - Busy installs with one ingest host:
scripts/nano upgradewaits for Vector's buffers to empty before it restarts ingest, and with log senders still sending at full rate they may not empty in time. The upgrade then stops safely before changing ingest. Fixed in v2.0.29: the upgrade closes the ingest edge to new events while Vector drains, so senders keep sending and retry. See Busy single ingest host. - Migrations run automatically. New PostgreSQL and ClickHouse migrations apply during the upgrade.
nano v2.0.29
nano v2.0.29: a security release. A package extract no longer widens ClickHouse's network allow-list or resets its sizing, busy single ingest hosts upgrade without pausing senders, and image bundles work on registry installs.
nano v2.0.27
nano v2.0.27: disk protection on by default, ClickHouse storage tiering, egress controls, and updated libraries and base images.