Fix StringSliceCSV treating explicit empty string as a one-element list

[pujitha24]

Motivation:
Issue #6581 reports that setting compactor.enabled_tenants: ""
explicitly in config causes the compactor to skip every tenant,
instead of behaving like the unset/omitted case (all tenants
allowed). The compactor logs "skipping user because it is not owned
by this shard" for every real tenant.

Approach:
StringSliceCSV.Set (pkg/util/flagext/stringslicecsv.go), which
backs -compactor.enabled-tenants/-compactor.disabled-tenants and
several other CSV-list flags (alertmanager, ruler, store-gateway,
ingester, querier, distributor, ring, limits, etc.), implemented
Set as strings.Split(s, ","). Go's strings.Split("", ",")
returns [""], a one-element slice containing an empty string, not
an empty/nil slice. YAML unmarshaling for enabled_tenants: ""
calls Set(""), so EnabledTenants ends up as [""] with
len() == 1.

users.NewAllowedTenants (pkg/util/users/allowed_tenants.go) treats
len(enabled) > 0 as "only tenants in this list are allowed", so
with enabled == [""] only a tenant literally named "" is
allowed and every real tenant is rejected. This same
NewAllowedTenants helper backs the identical enabled/disabled
tenant list pattern in the compactor, alertmanager, ruler, and
store-gateway, so all four are affected identically by explicitly
empty tenant-list config, though issue #6581 only reports it for the
compactor.

Fix Set to special-case an empty input string by setting the value
to nil, matching the sibling SecretStringSliceCSV.Set, which
already special-cases s == "" this same way.

Validation:
go build ./...
go test ./pkg/util/flagext/... ./pkg/compactor/... ./pkg/util/validation/... ./pkg/util/users/... -tags "netgo slicelabels"
All pass, including new tests in pkg/util/flagext/stringslicecsv_test.go
covering direct Set("") and YAML unmarshaling of csv: "".

Before/after: with enabled_tenants: "", no tenant was ever
compacted by any compactor replica (every tenant matched
"not owned by this shard"/was filtered out before ownership was even
checked). After this fix, an explicitly empty enabled_tenants (or
disabled_tenants) behaves identically to the field being omitted:
all tenants are eligible, subject to normal ring sharding.

Fixes #6581

Signed-off-by: Pujitha Paladugu 10557236+pujitha24@users.noreply.github.com

[Copilot]

Pull request overview

This pull request fixes a configuration edge case where CSV-backed string-slice flags/YAML fields (notably compactor.enabled_tenants / -compactor.enabled-tenants) treated an explicitly empty string ("") as a one-element slice containing "", which inadvertently enabled allow-list mode for a nonexistent empty tenant and caused all real tenants to be skipped.

Changes:

• Update StringSliceCSV.Set to special-case s == "" by resetting the slice to nil instead of using strings.Split, aligning behavior with “unset/omitted” semantics.
• Add unit tests covering both direct Set("") behavior and YAML unmarshalling of csv: "".
• Add a CHANGELOG bugfix entry documenting the user-facing impact and resolution.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated no comments.

File	Description
pkg/util/flagext/stringslicecsv.go	Treat empty input string as an empty/unset list by setting the slice to nil before returning.
pkg/util/flagext/stringslicecsv_test.go	Add tests to ensure empty-string inputs don’t produce [""] and YAML "" unmarshals to an empty list.
CHANGELOG.md	Document the bugfix and its impact on enabled/disabled tenant CSV lists.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[pujitha24]

The failing checks here are not caused by this diff — flagging in case that is what is holding it up.

Both failures are amd64-only (arm64 passes) and in code this PR does not touch: TestPartitionCompactor_ShouldCompactOnlyUsersOwnedByTheInstanceOnShardingEnabledAndMultipleInstancesRunning, a ring-sharding timing test in pkg/compactor.

Happy to rebase to re-trigger CI if that would help, or to dig further if you think I have this wrong.

[SungJin1212]

Thanks for fixing it.

Two nits before merging:

• Reference the PR number (#7714) rather than the issue.
• The entry could also be trimmed to a sentence or two.

[SungJin1212]

lgtm

[pujitha24]

This is approved with green CI and no outstanding comments as far as I can tell — ready whenever you have a moment. Happy to rebase first if you'd like it freshened.

[SungJin1212]

Could you move it to the ## master / unreleased section?

[pujitha24]

Moved it to master / unreleased.

[SungJin1212]

@pujitha24
Can you get a rebase?