Files
cozystack/hack/check-opensearch-rd-secrets.bats
T
Aleksei Sviridkin f3c28e91c4 test(opensearch): pin the trust anchor, the name guard and the upgrade
Most of what this chart decides about TLS is invisible in a rendered
manifest until a cluster disagrees with it. The upgrade case is the
sharpest: a release that already runs operator-managed HTTP TLS must
keep working when the chart takes the CA over, and nothing about the
rendered output says whether it will.

The negative assertions carry as much as the positive ones. The trust
anchor must stay out of the resource map, since a name grant there
would outlive the platform's verdict on it, and a null tls value must
leave the chart rendering rather than aborting the whole release. Both
are properties that fail silently and only a test states them.

The bats guard covers what helm-unittest cannot see: which Secrets the
opensearch-rd definition exposes, so a key-bearing one cannot be added
to that surface without the guard going red.

Assisted-by: LLM
Signed-off-by: Aleksei Sviridkin <f@lex.la>
2026-09-11 01:23:22 +03:00

142 lines
6.9 KiB
Bash

#!/usr/bin/env bats
# Unit test: the opensearch ResourceDefinition hands tenants the key-free trust
# anchor and nothing else.
#
# The chart issues opensearch-<name>.http-ca (CA certificate AND CA private key,
# the latter under tls.key), opensearch-<name>-http-server and
# opensearch-<name>-dashboards-server (each a server certificate AND its private
# key). None may reach a tenant: read access to the CA key would let the holder
# issue certificates for anything rather than merely verify the server. What a
# client needs is ca.crt alone, delivered by the CA-extraction controller as the
# key-free <release>.tenant-ca projection and selected by LABEL rather than by
# name, because a name grant conveys whatever happens to occupy that name.
#
# Asserted against the RENDERED chart, not against cozyrds/opensearch.yaml on
# disk. templates/cozyrd.yaml globs cozyrds/* and emits each match through
# .Files.Get, so what ships is the render: reading one source file would miss a
# second file in that directory, and a dotfile the glob picks up. Helm ignores
# dotfiles only under templates/, so one here does ship. That is the whole reason
# this file exists alongside hack/check-opensearch-readme.bats, which asserts the
# same include shape by reading the one source file — narrower on purpose, and
# blind to exactly this.
#
# This lives in bats rather than helm-unittest because packages/system/
# opensearch-rd has no `test:` target, and hack/helm-unit-tests.sh skips any
# package without one. If that package ever grows one, this belongs there.
#
# spec.secrets is compared WHOLE rather than probed for known-bad names. Three
# ways this breaks are invisible to a name-by-name check:
#
# - a name added under a different spelling, or via a helper;
# - an include with NO resourceNames and NO matchLabels, which restricts
# nothing;
# - a SECOND matchLabels include selecting a label the key-bearing Secrets also
# carry. cert-manager stamps controller.cert-manager.io/fao on every Secret it
# creates, all three of these among them, so one such entry grants the CA
# private key while every name in the file stays correct.
REPO_ROOT="$(cd "$(dirname "${BATS_TEST_FILENAME:-$0}")/.." && pwd)"
CHART="$REPO_ROOT/packages/system/opensearch-rd"
# The complete tenant-visible Secret surface. The credentials Secret by name; the
# trust anchor by label. Nothing else.
EXPECTED_SECRETS='{"exclude":[],"include":[{"resourceNames":["opensearch-{{ .name }}-credentials"]},{"matchLabels":{"internal.cozystack.io/tenant-ca":"true"}}]}'
# Order within include is meaningless -- the matcher ORs across entries -- so the
# array is sorted before comparing, and a missing exclude is normalized to the
# empty list it means. Neither can hide a grant: sort preserves every entry (it is
# not unique), and exclude only ever narrows.
#
# Kept as a filter string rather than a shell function because hack/cozytest.sh
# inserts `return 0` before any line that is exactly `}`. A helper whose own last
# command decides the outcome returns 0 regardless -- and whether a wrapper
# propagates a failure at all then depends on the shell running it. Repeating the
# helm invocation inline in each test costs two lines and owes nothing to that.
NORMALIZE='{exclude: (.exclude // []), include: (.include | sort)}'
# Every document count below uses `yq eval-all`, never `yq eval`. `eval` runs the
# expression once PER DOCUMENT, so `[.] | length` reports 1 per document rather
# than the number of documents. `eval-all` folds the stream into one expression.
# Empty input still yields 1, since yq synthesizes a null document, so emptiness
# is checked separately before counting.
@test "opensearch-rd renders exactly one ApplicationDefinition" {
# stderr is deliberately not discarded: when the render breaks, helm's own
# message is the only thing that says why.
rendered=$(helm template opensearch-rd "$CHART") || return 1
# Fail closed: an empty render makes every assertion below vacuous, and yq
# cannot distinguish it from one document.
if [ -z "$rendered" ]; then
echo "helm template produced no output for $CHART -- guard is blind." >&2
return 1
fi
count=$(printf '%s\n' "$rendered" | yq eval-all '[.] | length' -) || return 1
if [ "$count" != "1" ]; then
echo "Expected exactly 1 rendered document from $CHART, got: $count" >&2
echo "templates/cozyrd.yaml globs cozyrds/*, so every file there -- including" >&2
echo "a dotfile -- ships its own tenant grants. Extend this suite before" >&2
echo "adding one." >&2
return 1
fi
}
@test "opensearch-rd selects the key-free tenant-CA projection by label" {
rendered=$(helm template opensearch-rd "$CHART") || return 1
if [ -z "$rendered" ]; then
echo "helm template produced no output for $CHART -- guard is blind." >&2
return 1
fi
count=$(printf '%s\n' "$rendered" | yq eval-all \
'[.. | select(has("secrets")) | .secrets.include[] | select(.matchLabels."internal.cozystack.io/tenant-ca" == "true")] | length' \
-) || return 1
if [ "$count" -lt 1 ]; then
echo 'opensearch-rd does not select internal.cozystack.io/tenant-ca: "true"' >&2
echo "Without it the tenant cannot read ca.crt and TLS verification is impossible." >&2
return 1
fi
}
# Whole-structure comparison. Anything added to the tenant's Secret surface -- by
# name, by label, or by an unrestricted entry -- changes this string.
@test "opensearch-rd exposes no Secret beyond the credentials and the trust anchor" {
rendered=$(helm template opensearch-rd "$CHART") || return 1
if [ -z "$rendered" ]; then
echo "helm template produced no output for $CHART -- guard is blind." >&2
return 1
fi
# eval-all again: with `eval`, a second rendered document would emit a second
# JSON object and the comparison below would run on a concatenation.
actual=$(printf '%s\n' "$rendered" \
| yq eval-all --output-format=json '[.. | select(has("secrets")) | .secrets]' - \
| jq --sort-keys --compact-output "if length == 1 then .[0] else . end | $NORMALIZE") || return 1
expected=$(printf '%s' "$EXPECTED_SECRETS" \
| jq --sort-keys --compact-output "$NORMALIZE") || return 1
# Fail closed: an empty or null result means spec.secrets stopped being readable
# at this path, so the guard is no longer observing what it claims.
if [ -z "$actual" ] || [ "$actual" = "null" ]; then
echo "spec.secrets is missing or unreadable in the rendered ResourceDefinition -- guard is blind." >&2
return 1
fi
if [ "$actual" != "$expected" ]; then
echo "Unexpected tenant Secret surface in the rendered opensearch-rd spec.secrets:" >&2
echo " actual: $actual" >&2
echo " expected: $expected" >&2
echo "opensearch-<name>.http-ca holds the CA PRIVATE KEY, and -http-server and" >&2
echo "-dashboards-server each hold a server private key. None may be granted to" >&2
echo "a tenant, by name or by a label they also carry -- cert-manager stamps its" >&2
echo "own labels on all three." >&2
return 1
fi
}