fix: apply custom dedup rules to pulled alerts (closes #5370)

[jyoti369]

Summary

Fixes #5370 - Prometheus (and all other) providers now respect custom deduplication rules when pulling alerts.

Root Cause

Custom dedup rules (fingerprint_fields) were applied in format_alert() (webhook path) but not in get_alerts() (pull path). When alerts are pulled via the provider's _get_alerts() method, the fingerprint from the source system was used as-is, ignoring any user-configured fingerprint_fields.

For Prometheus specifically, the native fingerprint is based on alertname alone, so alerts with different labels (e.g. env_dc, group) were incorrectly matching as duplicates:

echo -n 'NodeLowDiskSpace4Hours' | sha256sum
6211e97de90695be41b49c773306001204872a6ebb36f4aa72a5bbd998ce7ac5

Fix

Added custom dedup rule application to BaseProvider.get_alerts(), mirroring the existing logic in format_alert(). This means ALL providers benefit from the fix when pulling alerts, not just Prometheus.

Changed files:

• keep/providers/base/base_provider.py - Apply get_custom_deduplication_rule + get_alert_fingerprint in get_alerts() after setting provider metadata
• tests/test_get_alerts_custom_dedup.py - Unit tests verifying custom dedup rules are applied to pulled alerts

Testing

Added 3 unit tests:

1. Custom dedup rule overwrites native fingerprint (different labels produce different fingerprints)
2. No custom rule preserves original fingerprint
3. Dot-notation field access works for nested label keys (matches issue reporter's labels.alertname, labels.env_dc, labels.group config)

---

Note

Medium Risk
Changes how alert fingerprints are computed for all pulled alerts when a tenant has custom dedup rules, which can affect deduplication/grouping behavior across the system. Scope is small and covered by new unit tests, but behavior changes may impact existing alert correlation expectations.

Overview
Pulled alerts returned by BaseProvider.get_alerts() now respect tenant custom deduplication rules by recalculating each alert’s fingerprint using the configured fingerprint_fields (mirroring the existing webhook format_alert() behavior).

Adds unit tests to ensure a custom rule overwrites the source/native fingerprint, the original fingerprint is preserved when no rule exists, and dot-notation field paths (e.g. labels.env_dc) work when building the fingerprint.

^{Reviewed by Cursor Bugbot for commit 492a69e. Bugbot is set up for automated code reviews on this repo. Configure here.}

[shahargl]

lgtm

[shahargl]

@jyoti369 UTs fails:

=========================== short test summary info ============================
FAILED tests/test_get_alerts_custom_dedup.py::TestGetAlertsCustomDedup::test_custom_dedup_rule_overwrites_fingerprint - TypeError: Can't instantiate abstract class BaseProvider with abstract methods dispose, validate_config
FAILED tests/test_get_alerts_custom_dedup.py::TestGetAlertsCustomDedup::test_custom_dedup_with_dot_notation_fields - TypeError: Can't instantiate abstract class BaseProvider with abstract methods dispose, validate_config
FAILED tests/test_get_alerts_custom_dedup.py::TestGetAlertsCustomDedup::test_no_custom_rule_keeps_original_fingerprint - TypeError: Can't instantiate abstract class BaseProvider with abstract methods dispose, validate_config
====== 3 failed, 935 passed, 83 skipped, 13 warnings in 166.56s (0:02:46) ======

[jyoti369]

E2E failure in run-mysql-with-redis is unrelated to this PR -- it's a flaky Playwright test (test_search_by_cel[contains for nested property]) that expected 10 rows but got 20. Our PR only touches base_provider.py dedup logic and its unit test, nothing in the E2E search/filter path.

47 E2E tests passed, just this one flaked. Updating branch now to re-trigger CI.

[jyoti369]

@jyoti369 UTs fails:

=========================== short test summary info ============================
FAILED tests/test_get_alerts_custom_dedup.py::TestGetAlertsCustomDedup::test_custom_dedup_rule_overwrites_fingerprint - TypeError: Can't instantiate abstract class BaseProvider with abstract methods dispose, validate_config
FAILED tests/test_get_alerts_custom_dedup.py::TestGetAlertsCustomDedup::test_custom_dedup_with_dot_notation_fields - TypeError: Can't instantiate abstract class BaseProvider with abstract methods dispose, validate_config
FAILED tests/test_get_alerts_custom_dedup.py::TestGetAlertsCustomDedup::test_no_custom_rule_keeps_original_fingerprint - TypeError: Can't instantiate abstract class BaseProvider with abstract methods dispose, validate_config
====== 3 failed, 935 passed, 83 skipped, 13 warnings in 166.56s (0:02:46) ======

fixed this, please re-check.

[jyoti369]

@shahargl All changes are done from my side. The E2E failure (mysql-with-redis) is flaky tests unrelated to this PR - our changes only touch base_provider.py and a test file. CI has been re-triggered. Ready for re-review when you get a chance.

[jyoti369]

@shahargl tests are fixed and CI has been re-triggered. Ready for re-review when you get a chance!