Skip to content

Commit 4ba9270

Browse files
committed
docs(tenant-application-logs): scope by stream filters, allowlist picker params, bound step
Use extra_stream_filters for the application scope (the identity fields are stream fields), add each picker endpoint's own parameters to the allowlist under the forced scope, clamp step so hits and stats_query_range cannot return an unbounded number of buckets, and note that the separate Merge_Log_Key fix also protects the application scope from a forged kubernetes.labels. Assisted-by: LLM Signed-off-by: IvanHunters <[email protected]>
1 parent 671cf85 commit 4ba9270

1 file changed

Lines changed: 5 additions & 5 deletions

File tree

  • design-proposals/tenant-application-logs

‎design-proposals/tenant-application-logs/README.md‎

Lines changed: 5 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -109,13 +109,13 @@ extra_stream_filters={kubernetes_namespace_name="<ns>"}
109109
On the application root, the proxy additionally scopes to one application by its **identity labels** (the same labels as #94, `apps_cozystack_io_application_{group,kind,name}`, the lineage keys sanitized as kube-state-metrics sanitizes them), set on every line at collection time by §6:
110110

111111
```text
112-
extra_filters=apps_cozystack_io_application_kind:=<Kind> apps_cozystack_io_application_name:=<name>
112+
extra_stream_filters={apps_cozystack_io_application_kind="<Kind>",apps_cozystack_io_application_name="<name>"}
113113
```
114114

115-
- Because the §6 identity labels and the application endpoint ship together (Phase 2), the application scope is a plain equality match, with no pod lookup. Lines from before Phase 2 do not carry the labels; they stay reachable through the namespace endpoint, which Goals and Failure cases treat as the fallback.
115+
- Because the §6 identity labels and the application endpoint ship together (Phase 2), the application scope is a plain equality match, with no pod lookup. The identity fields are stream fields (added to `_stream_fields` in §6), so the scope uses `extra_stream_filters`, the same mechanism as the namespace scope and what VictoriaLogs indexes on. Lines from before Phase 2 do not carry the labels; they stay reachable through the namespace endpoint, which Goals and Failure cases treat as the fallback.
116116
- `<plural>` must be a kind of `apps.cozystack.io`, `<name>` a DNS-1123 subdomain; values are quoted. Anything else is `400`. The proxy maps `<plural>` to the title-cased lineage `Kind` by reading the ApplicationDefinitions (the same mapping as #94), in one place, and uses the mapped value in `apps_cozystack_io_application_kind`, never the raw plural.
117-
- The upstream request is built from scratch: only allowlisted parameters (`query`, `start`, `end`, `limit` (clamped), `step`, and on `tail` only `since` or `start_offset`, both capped) and no client headers. VictoriaLogs takes the tenant from `AccountID` / `ProjectID` headers; the proxy always sets them. Client `extra_filters` / `extra_stream_filters` are dropped. `/select/tenant_ids` and every non-allowlisted path are not proxied.
118-
- **Bounded time window.** For `query`, `hits`, `stats_query_range`, and the picker endpoints the proxy requires or derives a finite `[start, end]` and clamps the maximum span; VictoriaLogs otherwise uses the oldest stored timestamp when `start` is omitted and would scan the whole retention. Inverted ranges are rejected with `400`. The `tail` resume cap (§5) is a separate bound.
117+
- The upstream request is built from scratch: only allowlisted parameters (`query`, `start`, `end`, `limit` (clamped), `step`, each picker endpoint's own parameters (`field` for `field_values` and `stream_field_values`, and the `facets` options), and on `tail` only `since` or `start_offset`, both capped) and no client headers. VictoriaLogs takes the tenant from `AccountID` / `ProjectID` headers; the proxy always sets them. Client `extra_filters` / `extra_stream_filters` are dropped. `/select/tenant_ids` and every non-allowlisted path are not proxied.
118+
- **Bounded time window.** For `query`, `hits`, `stats_query_range`, and the picker endpoints the proxy requires or derives a finite `[start, end]` and clamps the maximum span; VictoriaLogs otherwise uses the oldest stored timestamp when `start` is omitted and would scan the whole retention. It also clamps `step` so that `hits` and `stats_query_range`, which return one bucket per `step`, cannot produce an unbounded number of buckets over the span (the same cap on points as #94). Inverted ranges are rejected with `400`. The `tail` resume cap (§5) is a separate bound.
119119
- Queries using `options(ignore_global_time_filter=true)` are rejected, detected by a plain substring match on `ignore_global_time_filter` anywhere in `query`, subqueries included: it disables the time bounds the proxy relies on, and on `tail` it rescans the whole retention every tick.
120120
- Platform sources are out of reach: kube-apiserver audit and Kubernetes events carry no `kubernetes_namespace_name` in the current pipeline, so the stream filter excludes them.
121121

@@ -134,7 +134,7 @@ Resolve a namespace to its `VLCluster` with **exactly the rule the collector use
134134

135135
Fluent Bit keeps the lineage labels `apps.cozystack.io/application.{group,kind,name}` as `apps_cozystack_io_application_{group,kind,name}` (a second `nest` lift on `kubernetes_labels`, renamed before `Remove_wildcard kubernetes_`, like the three kept fields today) and adds them to `_stream_fields`. These give the application scope of §3 its labels, with the same names as #94. The lineage webhook sets the labels on pods whose owning application it resolves. This is the Phase 2 collection change.
136136

137-
**Related forgery fix, shipped separately.** `Merge_Log On` with no `Merge_Log_Key` lets a pod's own JSON log line add top-level keys to the record: a line such as `{"kubernetes":{"namespace_name":"<other>"}}` yields a second `kubernetes` map that the `nest` lift lifts next to the real one, so a forged `kubernetes_namespace_name` can route a line into another tenant's store that that tenant's Grafana already shows today. This is live in cozystack independent of this proposal, so it deserves its own fix: set `Merge_Log_Key` (for example `Merge_Log_Key log_processed`) so a parsed line lands under one key and cannot add top-level keys. That fix carries a breaking field-schema note: fields of JSON log lines searched at the top level today (for example `level`) become `log_processed.<field>`; in-repo charts do not read them, but any external dashboard or LogsQL query that does must migrate to `log_processed.<field>`. Phase 1 of this proposal depends on that fix having landed; a test asserts a crafted line cannot appear under another namespace.
137+
**Related forgery fix, shipped separately.** `Merge_Log On` with no `Merge_Log_Key` lets a pod's own JSON log line add top-level keys to the record: a line such as `{"kubernetes":{"namespace_name":"<other>"}}` yields a second `kubernetes` map that the `nest` lift lifts next to the real one, so a forged `kubernetes_namespace_name` can route a line into another tenant's store that that tenant's Grafana already shows today. The same way, a line setting `kubernetes.labels` could forge another application's identity labels, so this fix also protects the application scope above. This is live in cozystack independent of this proposal, so it deserves its own fix: set `Merge_Log_Key` (for example `Merge_Log_Key log_processed`) so a parsed line lands under one key and cannot add top-level keys. That fix carries a breaking field-schema note: fields of JSON log lines searched at the top level today (for example `level`) become `log_processed.<field>`; in-repo charts do not read them, but any external dashboard or LogsQL query that does must migrate to `log_processed.<field>`. Phase 1 of this proposal depends on that fix having landed; a test asserts a crafted line cannot appear under another namespace.
138138

139139
## User-facing changes
140140

0 commit comments

Comments
 (0)