[zzylol]

Why

#1151's three PRs (#1152, #1153, #1154) were merged into feat/1095-stage3, not main. #1144 (#1095's group MILP) builds on them through #1157. Landing them on main lets #1144 be rebased onto main and merged there, instead of keeping two long-lived branches.

What

The three reviewed commits, rebased onto main, plus one fix-up commit:

• docs: refresh the ingestion-time views of one refresh window in one run (#1151) #1152 (docs): the ingestion-time views of one refresh window are refreshed in one run (docs/design_docs/physical-planning-and-deployment.md, "Refresh groups").
• feat: derive ingestion refresh groups and refresh each group in one run (#1151) #1153: PhysicalASAPDAG::refresh_window / ingestion_refresh_groups; plan_installable's runtime refreshes each refresh group in one ingestion run.
• feat: charge a shared ingestion node once per refresh window (#1151) #1154: asap_cost::execution_demand::computations charges a NotMaterialized ingestion-time node once per distinct refresh window of its MatAtIngestion consumers, not once per consumer.
• Fix-up: the three commits predate Rename PaneChain to SummaryTumblingWindowChain, and every pane name to the tumbling-window wording #1163's rename. Their uses of PaneChain, pane_width_ms, pane_count and newest_panes become SummaryTumblingWindowChain, tumbling_window_width_ms, tumbling_window_count and newest_tumbling_windows, plus rustfmt.

Behavior change on main: plans where one ingestion Scan feeds several views of the same refresh window become cheaper, because the Scan is charged once per window, not once per view. So Stage 3 can select differently. #1179 generalizes this into explicit shared-scan candidates.

How

See #1152–#1154 for the design and the reviews. The rebase onto main was textually clean. The only change beyond the three commits is the rename fix-up.

Key code interfaces added / modified

As in #1153 and #1154; unchanged by the rebase except for the #1163 names. Summary:

• PhysicalASAPDAG::refresh_window(node) -> Option<RefreshWindow>;
• RefreshWindow { width_ms, phase_ms }, with RefreshWindow::of_window;
• ingestion_refresh_groups;
• asap_cost::execution_demand::computations: the per-refresh-window charge.

Major algorithm changes

#1154's ingestion charge, as reviewed there. Before: one computation per MatAtIngestion consumer. After: one per distinct refresh window among those consumers, and a consumer with an unbounded window counts alone.

Before this PR

main charged a shared ingestion Scan once per view, and the runtime contract did not group ingestion refreshes by refresh window.

After this PR

main has refresh groups and the per-refresh-window charge, the base #1144 expects.

Verification

On the rebased branch:

• cargo check --workspace --all-targets (except asap-frontend-metricsql, whose test target fails to compile on main itself: lowering_semantics.rs:160 calls lower_promql_query_workload with one argument);
• cargo fmt --all --check;
• cargo test -p asap-types --lib (142) and --test types -- materialized_view (11);
• cargo test -p asap-cost (all passed);
• cargo test -p asap-plan-selection --lib (285);
• cargo test -p asap-planner --test planner (82, 3 ignored);
• cargo test -p asap-integration-tests --test integration -- materialization_e2e planner_layering (88, 21 ignored).

These ran on the rebase onto c83aa19. Re-rebasing onto 59e8825 (#1178, #1180) was clean; the workspace check and fmt were re-run there.