[FSTORE-2030] Add support for specifying lookback windows for PIT queries#583
Merged
Conversation
manu-sj
force-pushed
the
FSTORE-2030
branch
2 times, most recently
from
37944b9 to
ddb1106
Compare
Contributor
There was a problem hiding this comment.
Pull request overview
This PR updates the Hopsworks Feature Store user guides to document the new lookback option for Point-in-Time (PIT) joins, including uniform and per-Feature-Group modes, and adds cross-links between batch-data retrieval and training-data materialization.
Changes:
- Adds a new “Lookback window for PIT joins” section to the batch-data guide, including
Lookback/Lookbacksexamples and pruning guidance. - Adds a corresponding training-data section that references the batch-data lookback documentation and notes persistence of the resolved window.
Reviewed changes
Copilot reviewed 2 out of 2 changed files in this pull request and generated 8 comments.
| File | Description |
|---|---|
| docs/user_guides/fs/feature_view/batch-data.md | Adds the main lookback documentation for PIT joins, with uniform/per-FG shapes and pruning notes. |
| docs/user_guides/fs/feature_view/training-data.md | Adds a training-data-facing lookback section and links readers to the batch-data lookback reference. |
javierdlrm
requested changes
Jun 1, 2026
…ries https://hopsworks.atlassian.net/browse/FSTORE-2030 PIT joins on partitioned feature groups defeat partition pruning: the join predicate (`right_fg.event_time <= root_fg.event_time`) is a range rather than an equality, so the engine has to scan an unbounded slice of history on every read. As feature groups grow with daily ingestion this linearly inflates scan cost. Lookback windows expose an explicit upper bound on the historical depth a PIT join may reach. This repo carries the user-facing reference for the new parameter. The batch-data and training-data user guides describe the `lookback` argument on `FeatureView.get_batch_data` and `create_training_data`. The pages cover the two-bound `Lookback` value, the trade-off between partition-key mode (pushed down to file-listing) and event-time mode (engine-dependent), and the per-feature-group override carrier. Each mode is shown in both the instance form and the dict-equivalent form with concrete partition-column placeholders and literal dates. Reviewed-by: GitHub Copilot <Copilot@users.noreply.github.com> Reviewed-by: OpenAI Codex (GPT-5 via codex-plugin-cc 1.0.4) <codex@openai.com> Signed-off-by: Manu Sathyarajan Joseph <manu.joseph@logicalclocks.com> Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…ries https://hopsworks.atlassian.net/browse/FSTORE-2030 Rename the lookback API surface across the SDK, backend, and docs: the feature-view-level container is now Lookback (top-level wire field "lookback") and the per-feature-group bound is FeatureGroupLookback, with feature_group_lookbacks naming the per-FG override map. Lookback keys are canonically uppercase (EVENT_TIME, PARTITION_KEY) with case-insensitive validation. The backend now echoes the persisted lookback configuration on training dataset responses so the client no longer rehydrates it locally. The batch-data and training-data guides switch their examples to the renamed classes, the feature_group_lookbacks keyword, and the uppercase canonical keys. Signed-off-by: Manu Sathyarajan Joseph <manu.joseph@logicalclocks.com> Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
javierdlrm
approved these changes
Jun 2, 2026
manu-sj
added a commit
that referenced
this pull request
Jun 3, 2026
…T queries (#583) (#597) * [FSTORE-2030] Add support for specifying lookback windows for PIT queries https://hopsworks.atlassian.net/browse/FSTORE-2030 PIT joins on partitioned feature groups defeat partition pruning: the join predicate (`right_fg.event_time <= root_fg.event_time`) is a range rather than an equality, so the engine has to scan an unbounded slice of history on every read. As feature groups grow with daily ingestion this linearly inflates scan cost. Lookback windows expose an explicit upper bound on the historical depth a PIT join may reach. This repo carries the user-facing reference for the new parameter. The batch-data and training-data user guides describe the `lookback` argument on `FeatureView.get_batch_data` and `create_training_data`. The pages cover the two-bound `Lookback` value, the trade-off between partition-key mode (pushed down to file-listing) and event-time mode (engine-dependent), and the per-feature-group override carrier. Each mode is shown in both the instance form and the dict-equivalent form with concrete partition-column placeholders and literal dates. Reviewed-by: GitHub Copilot <Copilot@users.noreply.github.com> Reviewed-by: OpenAI Codex (GPT-5 via codex-plugin-cc 1.0.4) <codex@openai.com> * [FSTORE-2030] Add support for specifying lookback windows for PIT queries https://hopsworks.atlassian.net/browse/FSTORE-2030 Rename the lookback API surface across the SDK, backend, and docs: the feature-view-level container is now Lookback (top-level wire field "lookback") and the per-feature-group bound is FeatureGroupLookback, with feature_group_lookbacks naming the per-FG override map. Lookback keys are canonically uppercase (EVENT_TIME, PARTITION_KEY) with case-insensitive validation. The backend now echoes the persisted lookback configuration on training dataset responses so the client no longer rehydrates it locally. The batch-data and training-data guides switch their examples to the renamed classes, the feature_group_lookbacks keyword, and the uppercase canonical keys. --------- Signed-off-by: Manu Sathyarajan Joseph <manu.joseph@logicalclocks.com> Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Lookback window for PIT joinstofeature_view/batch-data.mdcovering the two modes, the dict and dataclass call shapes, partition pruning behavior, and the one-sided lower-only form.feature_view/training-data.mdso users hittingcreate_training_datafind the same reference.JIRA
FSTORE-2030
Test plan
Companion PRs
logicalclocks/hopsworks-ee→ branchFSTORE-2030logicalclocks/hopsworks-api→ branchFSTORE-2030logicalclocks/loadtest→ branchFSTORE-2030