danny0405 opened a new issue, #20576: URL: https://github.com/apache/hudi/issues/20576
### Bug Description **What happened:** Flink stream-read metrics parse timeline timestamps using the reader JVM's default timezone, even when the table uses `hoodie.table.timeline.timezone=UTC`. This can offset `issuedInstant`, `splitLatestCommit`, and their delay gauges by hours. Found while reviewing #20564. This is pre-existing: the earlier implementation also called the same parser and compared its result with `Instant.now()`. Switching to an injected `Clock` does not introduce or fix the problem. At reviewed commit `bc41326be5c24f8e400e67da8e3174ed8071c6d2`, [HoodieInstantTimeGenerator.parseDateFromInstantTime](https://github.com/apache/hudi/blob/bc41326be5c24f8e400e67da8e3174ed8071c6d2/hudi-common/src/main/java/org/apache/hudi/common/table/timeline/HoodieInstantTimeGenerator.java#L86) parses a `LocalDateTime` and unconditionally applies `ZoneId.systemDefault()`. `FlinkStreamReadMetrics` uses this parser without the table's timezone. **What you expected:** Parse the commit timestamp in the timezone used by the table timeline before comparing absolute instants. For a UTC timeline, changing the reader JVM's timezone must not change the epoch timestamps or measured delays. **Steps to reproduce:** 1. Use a table with `hoodie.table.timeline.timezone=UTC`. 2. Run the reader JVM with `-Duser.timezone=Asia/Shanghai`. 3. Consider commit `20261111100000000` (10:00 UTC), with current time fixed at `2026-11-11T10:10:00Z`. 4. The existing parser interprets the commit as 10:00 Shanghai time, or `02:00Z`. The calculated delay is 29,400 seconds instead of 600 seconds. Minimal parser-level reproducer, run with Hudi on the classpath and the JVM timezone above: ```java Instant parsed = HoodieInstantTimeGenerator .parseDateFromInstantTime("20261111100000000") .toInstant(); Instant now = Instant.parse("2026-11-11T10:10:00Z"); System.out.println(parsed); System.out.println(Duration.between(parsed, now).getSeconds()); ``` The output implied by the current implementation is `2026-11-11T02:00:00Z` and `29400`. This report is based on source inspection; the snippet and an end-to-end Flink job have not been executed. ### Environment **Hudi version:** Source reviewed at `bc41326be5c24f8e400e67da8e3174ed8071c6d2` in #20564; the parser behavior predates that PR. **Query engine:** Flink, legacy stream reader and Source V2 metric paths. **Relevant configs:** - Table: `hoodie.table.timeline.timezone=UTC` - Reader JVM: `-Duser.timezone=Asia/Shanghai` ### Suggested scope and validation - Obtain the table timezone via `metaClient.getTableConfig().getTimelineTimezone().getZoneId()` and pass it explicitly to metric timestamp parsing. - Consider a reusable `parseDateFromInstantTime` overload accepting `ZoneId`, preserving existing timestamp compatibility handling. Avoid changing shared JVM-wide timezone state to parse a particular table. - Test UTC timelines with non-UTC reader JVMs, using an independently fixed expected `Instant`. Building the test clock with the same parser can conceal the error. - Cover local timelines with matching writer/reader zones and retain idle/active split behavior. **Separate LOCAL limitation:** `LOCAL` resolves to the current JVM's default timezone; it does not persist the writer's concrete zone ID. Reading a LOCAL timeline in another timezone cannot reliably recover the original zone from this setting alone. Document or track that limitation separately from honoring an explicit UTC configuration. ### Logs and Stack Trace No exception is required; the failure is incorrect metric values. ### Related work - #15907 tracks the broader requirement for timeline APIs to honor the configured timezone. This issue scopes the Flink stream-read metrics follow-up. - #20564 fixes metric lifecycle and live delay reporting; this timezone fix can be handled independently. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
