nano v2.0.20
nano v2.0.20 redacts credentials in raw event text, limits which networks reach ClickHouse, sizes from daily volume and parses up to 2.6× faster.
v2.0.20 includes everything in v2.0.19.
To upgrade, follow Upgrade a self-hosted deployment. On Docker Compose, scripts/nano upgrade upgrades v2.0.19 to v2.0.20 directly. Earlier v2.x releases also upgrade directly; there are no required stops. Read Upgrade notes before you upgrade.
Security
- Credentials are redacted inside raw event text. nano already removed fields named like credentials before storage. A credential written inside a string was not covered, so a line such as
{"Authorization": "Bearer …"}could be kept inraw_dataandmessage. Now the value is replaced with[REDACTED]and the rest of the text is kept. This covers JSON keys, header lines (Authorization: Bearer …),key=valuepairs and cookies, where each cookie value is redacted and the cookie names stay. Text that names a credential but can't be safely bounded is replaced as a whole with[REDACTED: credential-bearing text].- The names covered are
authorization,proxy-authorization,x-api-key,x-webhook-secret,cookieandset-cookie. - It applies on every ingest lane: HTTP, HEC, OTLP logs, traces and metrics, pull sources, upstream Vector, parse-error records and the Tenzir direct OCSF pipeline.
- Rows already stored are not purged. Events stored before the upgrade keep their text.
- The names covered are
- ClickHouse users accept connections only from internal networks. Before, ClickHouse users accepted connections from any address, bounded only by the network around ClickHouse.
- Docker Compose: the installer limits every user to loopback and the Docker address pools the Compose network uses. See ClickHouse network allow-list.
- Multiple servers: every user accepts only loopback and the inventory's servers. See ClickHouse network allow-list.
- Helm: set
clickhouse.allowedNetworks. Until you set it, the previous any-address behavior stays, so no upgrade locks anything out.helm installandhelm upgradeprint a warning, and SIEM Health reports "ClickHouse users accept connections from any address". See ClickHouse network allow-list. - An external Tenzir or Cribl writer is admitted as the ingest user only. See the pages above for the setting.
- Raw-SQL users can't read sensitive system tables. ClickHouse denies the raw-SQL users the sensitive system tables, such as
system.query_logandsystem.users. nano also rejectssystem.*in raw SQL. Each nano user now holds grants for only the system tables it reads.
Capacity
-
Size from your planned volume. Set
NANO_PLANNED_GB_PER_DAYon Docker Compose, or pick a Helm preset (small,standard,large,xlarge). From about 1 TB/day, nano applies the large-deployment ClickHouse and Vector settings: higher part limits, adaptive insert flushes and more Vector sink concurrency. The planned-volume tiers change nothing unless you set it. See Plan for your daily volume and Sizing presets. -
Vector sizing uses events per second. Parsing costs about the same per event, whatever its size, so a TB of small events needs more Vector CPU than a TB of large ones. Set your average event size with
NANO_AVG_EVENT_BYTES(Compose) orsizing.avgEventBytes(Helm). The default is 1000. For raw logs parsed by nano, that works out to:Average event size Typical feeds Vector vCPU per TB/day About 300 B Firewall, flow, DNS About 41 1,000 B (default) Mixed, identity, cloud audit About 13 About 2,000 B EDR About 7 Pre-shaped OCSF from Tenzir or Cribl needs about 0.5 vCPU per TB/day.
-
Vector memory floor. On Docker Compose, the installer aims to give Vector at least 1 GiB of memory, with or without a planned volume. It is best effort: the memory comes from PostgreSQL and the app services, never ClickHouse, and a small host can't always reach it. See Upgrade notes.
-
Prevalence refresh scales with volume. The rare and new lists refresh every 10 minutes by default, every 30 minutes at about 1 TB/day and every 60 minutes at 10 TB/day and up. At those volumes the lists can lag ingest by up to one refresh. Detections still check a new artifact against the real-time counts, so it is still caught as rare.
-
Shard balance in SIEM Health. On multi-shard ClickHouse, SIEM Health reports "ClickHouse shards are out of balance" when one shard's insert rate, part count or merge load stays well away from the others. It names the shard and the metric. If the check can't run, SIEM Health says so.
-
Tenzir writes to every shard. On a multi-shard cluster, the Tenzir pipeline writes to the distributed table, so events spread across shards. Helm publishes the shard count and table in a
<release>-direct-ocsfConfigMap for an external writer.
Performance
- Parsing is up to 2.6× faster per event. Across 16 benchmarked shipped parsers the gain is 1.07–2.6×. The EDR sources gain the most, 1.75–2.6×: CrowdStrike Falcon 2.6×, Microsoft Defender for Endpoint 2.0×, SentinelOne 1.9× and Sysmon 1.75×.
- Part of the gain is a parser update. To get it, re-sync your parser repository, run Apply upstream update on each log source, then deploy. Applying the update alone doesn't change what Vector runs.
OCSF
- Failed parses are searchable. An event a parser can't handle is stored as an OCSF parse-error record (
class_uid=0), with the original text inraw_dataand the reason inunmapped.parse_error. Search for them by log source, for examplemetadata.log_source="my_source" class_uid=0. - More pages read OCSF. The log-source test lab, Nano Pipe preview, hunt evidence and scoring, and shadow investigation now read OCSF events. Before, they could come back empty on OCSF installs.
- The Audit page reads both stores. Audit entries from before and after the upgrade are listed. Platform audit events, and case audit events written after the upgrade, show once. Case audit entries from before the upgrade stay listed under their original ids, and their OCSF copies are hidden. Known edge: when a search window starts in the moment (about 80 ms) between an older case entry and its OCSF copy, that entry shows only when you widen the window.
- pivt writes OCSF queries. pivt's prompts use OCSF field names, and its query checks run against your install's schema. This stops wasted retries and false "Query may have issues" warnings.
protocol="tcp"works. Protocol names such astcp,udpandicmpnow match OCSF's protocol numbers.- Legacy field names say what to use instead.
host,hostname,src,dest,signature,appanddest_domainused to match nothing on OCSF. A search that uses them now returnsdid you mean <OCSF path>. A detection rule that uses them is refused when you save it. Stored rules that use them fail and back off for an hour instead of retrying every minute. A field you create withevalunder one of those names still works.
Fixes
| whereon an array field after| tablereturned no rows. This was broken since v2.0.18. Searches on array fields aftertable,renameorevalcopies now test membership again, andstats bysplits arrays into single values.- Two-shard multi-server installs. Refresh activation now works on both shards.
- Case audit events showed twice on OCSF installs. New case audit events now show once.
Upgrade notes
- Upgrade with the release, not by editing image tags. On Docker Compose, use
scripts/nano upgradeor re-runinstall.shfrom the new bundle. On Kubernetes, usehelm upgradewith the new chart. These deliver the new ClickHouse user and config files. An install that only changes image tags keeps its old ClickHouse users: they stay open to any address, and some pages, such as tiering stats, fail with an access error. - Helm: set
clickhouse.allowedNetworks. Until you do, ClickHouse users accept connections from any address and SIEM Health reports it. See ClickHouse network allow-list. - Some audit entries change id. After the upgrade, existing platform audit entries that also have an OCSF copy are listed under their OCSF ids, so their ids exported or bookmarked before the upgrade won't match. Case audit entries from before the upgrade keep their original ids.
- Vector memory may move. On Docker Compose, the installer now aims to give Vector at least 1 GiB, even without a planned volume. On upgrade, it can take memory from PostgreSQL or the app services to do it. A small host, around 4 GiB of RAM, can't reach 1 GiB. There, pin
VECTOR_MEM_LIMITbelow the managed block in.env, or use a larger host. See Plan Vector from events per second. - Old parse failures stay where they were. Parse failures from before the upgrade aren't moved; only new ones are stored as OCSF parse-error records.
- Re-deploy parsers for the speed-up. Re-sync the parser repository, run Apply upstream update, then deploy each log source.
- Review rules that use legacy field names. Rules that use
host,hostname,src,dest,signature,appordest_domainmatched nothing on OCSF. After the upgrade they fail and back off. Change them to the OCSF path in the error message.
nano v2.0.22
nano v2.0.22 hotfix: large parsers such as Sysmon no longer roll back on small hosts, and Helm installs keep ClickHouse reachable with allowedNetworks.
nano v2.0.19
nano v2.0.19 fixes distinct counts that counted a missing field, keeps every key of HEC object events and improves the Kubernetes guide for GKE and EKS.