osscm commented on code in PR #11041: URL: https://github.com/apache/iceberg/pull/11041#discussion_r4098811303
########## format/view-spec.md: ########## @@ -160,7 +180,116 @@ Each entry in `version-log` is a struct with the following fields: | _required_ | `timestamp-ms` | Timestamp when the view's `current-version-id` was updated (ms from epoch) | | _required_ | `version-id` | ID that `current-version-id` was set to | -## Appendix A: An Example +#### Storage Table Identifier + +The table identifier for the storage table that stores the precomputed results. + +| Requirement | Field name | Description | +|-------------|----------------|-------------| +| _required_ | `namespace` | A list of strings for namespace levels | +| _required_ | `name` | A string specifying the name of the table | + +### Storage table metadata + +This section describes additional metadata for the storage table that supplements the regular table metadata and is required for materialized views. +The `refresh-state` property is set on the [snapshot summary](https://iceberg.apache.org/spec/#snapshots) property of a storage table snapshot to provide information about the state of the precomputed data. + +| Requirement | Field name | Description | +|-------------|-----------------|-------------| +| _optional_ | `refresh-state` | A [refresh state](#refresh-state) record stored as a JSON-encoded string | + +#### Freshness + +A materialized view is **fresh** when the storage table represents the result of the current view query. However, consumers may still decide to consume from a stale storage table based on their own policies. + +A change to the materialized view's definition produces a new `view-version-id`; any storage-table snapshot recorded at a prior `view-version-id` is invalid and should not be consumed until refreshed. + +#### Refresh state + +The refresh state record captures the state of dependencies that the producer chose to track from the materialized view's dependency graph. A dependency is recorded in `source-states` as either a `table` entry (a source table or an source materialized view's storage table) and/or a `view` entry. Source materialized views can be stored as a `view` and a `table` entry. + +The refresh state has the following fields: + +| Requirement | Field name | Description | +|-------------|------------------------------|-------------| +| _required_ | `view-version-id` | The `version-id` of the materialized view when the refresh operation was performed | +| _required_ | `source-states` | A list of [source state](#source-state) records capturing the dependencies the producer chose to track; may be empty | +| _required_ | `refresh-start-timestamp-ms` | A timestamp of when the refresh operation was started | + +##### Producer: Recording Refresh State + +Producers may selectively choose a subset of their dependencies to record — for example, skipping non-Iceberg sources or recording an empty list. See [Appendix B](#appendix-b-what-counts-as-a-dependency) for strategies on how to store dependency state. + +When writing the refresh state, producers: + +* **Must** record `view-version-id` and `refresh-start-timestamp-ms`. +* **Should** include all distinct source states for the inputs they chose to track if they are reachable through multiple paths in the dependency graph. +* **May** leave `source-states` empty (e.g., when sources are non-Iceberg or freshness is determined by a mechanism outside this spec). + +##### Consumer: Evaluating Refresh State + +Consumers may use any combination of the following to assess the state of dependencies used to produce the storage table. + +* **Recency policy.** Accept the storage table when `refresh-start-timestamp-ms` falls within a staleness window. A recency policy bounds data age but does not establish freshness. +* **Trust the recorded `source-states`.** Compare each entry against the current catalog state — `snapshot-id` for tables, `version-id` for views, optionally recursive verification for source materialized views recorded by their storage tables. Also confirm that the recorded `view-version-id` equals the materialized view's current `view-version-id`. Review Comment: Question for the group - context for why we're watching this spec closely: we filed a feature request (StarRocks/starrocks#78193) to mirror StarRocks materialized views into an Iceberg catalog, with a later phase explicitly targeting this spec's storageTable/RefreshState once it stabilizes. So getting the freshness semantics right here matters for our roadmap. Is a raw snapshot-id comparison here safe against compaction/rewrite (replace) snapshots on the source table? If continuous compaction runs on the source table after the MV's last refresh - common in production - the table's current snapshot-id advances even though no logical row data changed. A recorded snapshot-id != current snapshot-id check would then report the MV as stale, potentially triggering an unnecessary full re-evaluation even though nothing user-visible changed. This isn't hypothetical - the same failure mode (full-partition refresh triggered by Iceberg compaction rather than actual data changes) has been reported against StarRocks' own native incremental-refresh path scanning Iceberg source tables. Is that risk already ruled out under the current wording here - e.g. an expectation that consumers key off the source table's last data-changing sequence-number rather than raw snapshot-id (feels adjacent to the sequence-number point @JanKaul raised earlier in the thread)? Genuinely not sure if this is a known non-issue or worth calling out explicitly - curious what others think. -- 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] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
