stack-593 - Release Notes

Release Date: 2026-06-14

Summary

Stack-593 is a stability and search infrastructure release, with a significant platform advancement in how portal search indexes are queried. A total of 13 tickets were completed across the PLFM and SWC projects.

The most impactful change for portal developers and search power users is a major upgrade to the Search Index query API: rather than being limited to a narrow, predefined set of filter and aggregation types, callers can now send any OpenSearch-native query expression (with appropriate safety guardrails). This unlocks compound queries, custom aggregations, and more expressive data exploration — all without requiring new API schema fields for each use case. Alongside this, a critical fix resolves large search indexes getting permanently stuck in a "creating" state, which had been blocking portal data from becoming searchable.

The Curator grid AI assistant also receives a reliability fix for timeout errors that were causing chat interactions to fail. CSV upload to Synapse Tables gets a modernized experience with a fully rewritten upload dialog. All Synapse users benefit from improved search URL continuity — existing bookmarks and links to the old Search page now redirect correctly to the updated search experience. Several infrastructure and platform stability improvements round out the release.

Affected Users & Systems

Deprecated or Breaking Changes

New Features

Search Index: Full OpenSearch Query Language Support New

The Search Index query API has been significantly expanded. Previously, only a narrow subset of query and filter types were supported — adding each new capability required new API schema fields and custom translation logic. Now, portal developers can send any OpenSearch-native query expression directly, including compound queries, custom aggregations, suggesters, and deep-pagination cursors. The platform validates the query for safety (blocking scripts, cross-index access, and overly complex queries) and transparently maps user-facing column names to internal identifiers. This makes the Search Index far more flexible for data exploration use cases without requiring ongoing platform changes for each new query pattern.

Who benefits: Portal developers, data scientists, and teams building search-powered experiences on Synapse who need expressive querying beyond simple keyword and filter searches.

Redesigned CSV Upload Experience for Synapse Tables New

The wizard used to upload CSV files into Synapse Tables has been fully rebuilt with a modern, React-based interface. The new experience handles both creating a new table from a CSV and appending rows to an existing table, with a cleaner preview step and more reliable handling of backend operations. The legacy upload wizard has been retired.

Who benefits: All Synapse users who create or update Tables by uploading CSV files.

Old Search URLs Now Redirect to Updated Search New

Links and bookmarks pointing to the old Synapse Search page now automatically redirect to the updated Search experience. If the URL contains a Synapse entity ID, the redirect goes directly to that entity's page; otherwise, users land on the new Search with the original query preserved where possible. This ensures no one gets a broken page from following an old link.

Who benefits: All Synapse users who have bookmarked search pages or received links to the old Search experience.

Fixes & Improvements

Large Search Indexes No Longer Get Stuck in "Creating" State

A critical bug caused search indexes containing tens of thousands of rows to get permanently stuck and never become usable. After documents were successfully written to the index, a propagation timing check would fail for large datasets, triggering repeated retries that ultimately dead-lettered the job and left the index frozen. The fix relaxes this timing check — the index now transitions to an active state and self-corrects as expected, rather than being abandoned. Portals with large datasets will now see their search indexes complete reliably.

Curator AI Assistant: Timeout Errors Resolved

The Curator's AI chat assistant was intermittently failing with a connection timeout when processing certain requests. This caused AI chat jobs to fail silently, leaving users without a response. The underlying network timeout handling has been fixed, and AI assistant requests now complete reliably.

Curator Grid: Multi-Value Filtering Now Works Correctly

Filtering grid rows by multiple values at once (e.g. showing only rows where a column matches any of several categories) was failing with an error and returning no results. Two related bugs have been fixed: one affecting filters that check for equality against a list of values, and one affecting "IN" filter operations. Multi-value filtering in the Curator grid now works as expected.

Portal Table Data Loading Failure Resolved

Table data was failing to load across several portals including ADKnowledge Portal, NF Portal, and Synapse, causing pages to hang indefinitely. The root cause — a table left in an invalid state after editing in the web client — has been identified and resolved. Affected tables can now be queried successfully.

NamHub Sign-In Loop Fixed

Users signing in to Synapse from the NamHub portal were experiencing a page refresh loop that prevented authentication from completing. The NamHub site was missing from Synapse's cross-origin allowlist, causing sign-in requests to be blocked. This has been corrected, and NamHub users can now sign in normally.

Internal: Deployment Pipeline Improvements

Several infrastructure-level fixes were made to improve the reliability of the release deployment process, including resolving a CloudFormation change set conflict and a Beanstalk application startup failure that affected some instances during the stack-592 rollout. Additionally, the internal migration utility now supports migrating data to the test stack environment, enabling more thorough pre-release validation. No user-facing impact.

Internal: Object Processing Queue Backlog Resolved

An internal processing queue that indexes platform objects was growing unexpectedly after the stack-592 deployment, indicating a backlog of unprocessed events. The root cause has been identified and resolved, restoring normal queue throughput. No user-facing impact.