[jira] [Updated] (YARN-11972) Support comma-separated usernames in Capacity Scheduler user placement rules

["Manish Maheshwari (Jira)" <ji...@apache.org>]

     [ https://issues.apache.org/jira/browse/YARN-11972?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Manish Maheshwari updated YARN-11972:
-------------------------------------
    Priority: Minor  (was: Major)

> Support comma-separated usernames in Capacity Scheduler user placement rules
> ----------------------------------------------------------------------------
>
>                 Key: YARN-11972
>                 URL: https://issues.apache.org/jira/browse/YARN-11972
>             Project: Hadoop YARN
>          Issue Type: Improvement
>            Reporter: Manish Maheshwari
>            Priority: Minor
>
> h2. Summary
> Support multiple discrete usernames in a *single* Capacity Scheduler placement rule by accepting a comma-separated list in the {{matches}} field (e.g. {{alice,bob,charlie}}), so administrators can route many named users to one queue without creating 1:1 rules or relying on LDAP group mapping.
> h2. Problem
> User-type placement rules currently match *exactly one username* (or {{*}} for all users). The {{matches}} value is passed verbatim to {{createUserMatcher()}} in {{MappingRuleCreator}}.
> Entering {{"matches": "a,b,c,d"}} only matches a user literally named {{a,b,c,d}} — not users {{a}}, {{b}}, {{c}}, or {{d}} individually.
> {{OrMatcher}} already exists in {{MappingRuleMatchers}} but is not wired into rule creation.
> h2. Proposed Solution
> When {{type=user}}, parse {{matches}} as a comma-separated list (trim whitespace):
> * *One username* → {{createUserMatcher(name)}} (unchanged)
> * *Multiple usernames* → {{OrMatcher}} over per-user matchers
> * *Entire string is {{*}}* → {{createAllMatcher()}} (unchanged; no split)
> Apply the same logic in legacy {{MappingRule.createLegacyRule()}} for {{u:}} mappings (e.g. {{u:alice,bob:queue1}}).
> *Out of scope (v1):* comma-separated groups, per-user wildcards, JSON array syntax, Fair Scheduler.
> h2. Acceptance Criteria
> *AC-1:* JSON rule with {{"matches": "alice,bob,charlie"}} matches users {{alice}}, {{bob}}, {{charlie}}; does not match {{dave}}.
> *AC-2:* Whitespace tolerance — {{" alice , bob "}} behaves same as {{"alice,bob"}}.
> *AC-3:* Single-user and wildcard unchanged. {{*}} as match-all applies only when the entire string is {{*}} (not {{*,alice}}).
> *AC-4:* Legacy {{yarn.scheduler.capacity.queue-mappings=u:alice,bob:queue1}} places {{alice}} and {{bob}} in {{queue1}}. Existing {{u:user1:queue1,u:user2:queue2}} (comma separates rules) unchanged.
> *AC-5:* Capacity Scheduler UI accepts comma-separated usernames in Match Pattern; help text updated.
> *AC-6:* CS UI round-trip — comma-separated list saved and reloaded as one rule row.
> *AC-7:* {{type=group}} and {{type=application}} — comma in {{matches}} remains literal (no split in v1).
> h2. Minimal Patch Outline
> All under {{hadoop-yarn-project/}}:
> *Phase 1 — RM engine (required)*
> # {{MappingRuleMatchers.java}} — add {{createUserMatcherFromMatches(String)}}
> # {{MappingRuleCreator.java}} — use new factory for USER case
> # {{MappingRule.java}} — same in {{createLegacyRule()}}
> # {{MappingRulesDescription.json}} — document comma-separated syntax
> # {{TestMappingRuleCreator.java}}, {{TestCSMappingPlacementRule.java}}, {{TestMappingRuleMatchers.java}} — tests
> *Phase 2 — CS UI + docs (recommended)*
> # {{PlacementRuleForm.tsx}} — update placeholder/help text
> # {{placement-rule-schema.ts}} — reject empty tokens after split
> # {{CapacityScheduler.md}} — document multi-user syntax
> h2. Edge Cases
> * Empty entry {{alice,,bob}} — reject at parse with clear error
> * Username containing comma — not supported in v1; document limitation
> * Very long lists (500+ users) — engine OK; UI usability separate concern
> h2. Alternatives Considered
> * UI expands one row → N JSON rules at save — works today; config bloat
> * Group mapping — not acceptable where explicit user lists are required
> * JSON array syntax — defer to v2



--
This message was sent by Atlassian Jira
(v8.20.10#820010)

---------------------------------------------------------------------
To unsubscribe, e-mail: yarn-issues-unsubscribe@hadoop.apache.org
For additional commands, e-mail: yarn-issues-help@hadoop.apache.org