Query fuzz tests: remove the #7550 duplicate-series error workaround once Prometheus >= v3.15.0 is vendored

[sandy2008]

AI Tool Usage Notice
If you used an AI tool to help draft this issue,
please make sure you have reviewed and validated all content before submitting.
You are responsible for the accuracy and quality of everything in this request.
Low-quality or unreviewed AI-generated submissions may be closed without further investigation.
See our Generative AI Contribution Policy for details.

Is your feature request related to a problem? Please describe.
This tracks the follow-up agreed in #7550. That PR added a tactical workaround to integration/query_fuzz_test.go for the flake in #7546. The PromQL engines emitted found duplicate series for the match group …: [X, Y] with the two series in non-deterministic order, so Cortex and the reference engine could report the same error as [X, Y] vs [Y, X]. The workaround has two parts:

• duplicateSeriesRE and canonicalizeDuplicateSeriesErr canonicalize that one message before comparing. sameErrorClass uses them at four sites: TestDisableChunkTrimmingFuzz, TestExpandedPostingsCacheFuzz, TestLazyMatchersFuzz and runQueryFuzzTestCases.
• TestSameErrorClass tests the helper.

As agreed in the #7550 review, it should be removed "once both upstream fixes are vendored", so the fuzz tests don't permanently encode upstream error-message formats. That is currently blocked.

Status as of 2026-09-25:

• execution/binary: make duplicate series error message deterministic thanos-io/promql-engine#711: merged 2026-05-29 and already vendored. Cortex's promql-engine 1b359e941bde includes it.
• promql: make duplicate series error message deterministic prometheus/prometheus#18810: merged into main on 2026-08-17. The first release that contains it is v3.15.0, currently at v3.15.0-rc.1. It is not in v3.14.0 or v3.13.3.
• Cortex: vendors Prometheus v0.309.1 (v3.9.1). images.Prometheus, the reference server used by TestPrometheusCompatibilityQueryFuzz, TestNativeHistogramFuzz, TestExperimentalPromQLFuncsWithPrometheus and TestStoreGatewayLazyExpandedPostingsSeriesFuzzWithPrometheus, is quay.io/prometheus/prometheus:v3.9.1.

Cortex can't simply bump Prometheus. Newer Prometheus releases no longer ship github.com/prometheus/prometheus/tsdb/errors; it is absent from both v3.14.0 and v3.15.0-rc.1. That package is imported by 6 Cortex files and by the vendored Thanos (pkg/block/writer.go). Building master against v0.315.0-rc.1 fails on the Thanos import. Thanos main is still on Prometheus v3.9.1, and its upgrade PR, thanos-io/thanos#8774 (to 3.11), is a draft with conflicts.

Describe the solution you'd like
Once Cortex vendors Prometheus ≥ v3.15.0, which requires a Thanos version that supports it:

1. Bump images.Prometheus in integration/e2e/images/images.go to ≥ v3.15.0.
2. Delete duplicateSeriesRE, canonicalizeDuplicateSeriesErr and TestSameErrorClass. Go back to a strict error comparison at the four call sites, as before fix(integration): compare query fuzz errors by class, not exact string #7550.
3. Keep a tolerance for TestBackwardCompatibilityQueryFuzz until the latest Cortex release includes the fix. That test compares against getLatestReleaseImage(), whose engine still emits the old order. Alternatively, do the removal in two steps: everything else first, and the backward-compatibility case one release later.

Both upstream fixes put the smaller rendered label set first, so strict comparison will also hold between the Prometheus and Thanos engines.

Describe alternatives you've considered

• Remove the workaround now: this brings back the roughly weekly flake from Flaky test: TestPrometheusCompatibilityQueryFuzz / TestExperimentalPromQLFuncsWithPrometheus — error-string comparator sensitive to non-deterministic series-list ordering #7546. Both the vendored engine and the v3.9.1 reference image still emit a non-deterministic order.
• Keep the workaround permanently: the fix(integration): compare query fuzz errors by class, not exact string #7550 review turned this down so the fuzz tests don't encode upstream error-message formats.
• Bump only Prometheus in Cortex: not possible while the vendored Thanos imports tsdb/errors.
• Get #18810 backported to 3.9.x: there has been no 3.9.x patch release since v3.9.1 (2026-01-07), so this isn't realistic.

Additional context

• Flake and discussion: Flaky test: TestPrometheusCompatibilityQueryFuzz / TestExperimentalPromQLFuncsWithPrometheus — error-string comparator sensitive to non-deterministic series-list ordering #7546, fix(integration): compare query fuzz errors by class, not exact string #7550.
• Upstream fixes: promql: "found duplicate series for the match group" error message is non-deterministic prometheus/prometheus#18809 / promql: make duplicate series error message deterministic prometheus/prometheus#18810, execution/binary: "found duplicate series for the match group" error is non-deterministic thanos-io/promql-engine#710 / execution/binary: make duplicate series error message deterministic thanos-io/promql-engine#711.
• Blocking chain:
  • the Prometheus v3.15.0 release;
  • Thanos moving past Prometheus v3.9.1 (Prometheus 3.11 upstream thanos-io/thanos#8774 or later);
  • a Cortex bump of github.com/prometheus/prometheus to ≥ v0.315.0, plus Thanos.