github hasura/graphql-engine v2.50.10

latest release: v2.51.9
one hour ago

Changelog

This is a patch release for v2.50.

Bug fixes and improvements

Server

  • Fix a segfault or corrupted results that could occur when gzip-ed responses larger than 2GB are returned.
  • Improve the performance for the common case of queries without fragments (Thanks to Lorenzo for the contribution)
  • run_sql executed with no_transaction: true now splits multi-statement SQL using a full lexical scanner. Semicolons inside ordinary and escape strings, quoted identifiers, dollar-quoted blocks, and line or block comments are no longer treated as statement separators, so dollar-quoted PL/pgSQL blocks, regular-expression anchors, positional parameters, and comments or strings that contain ; are executed as written instead of being mis-split.
  • HASURA_GRAPHQL_UNAUTHORIZED_ROLE is now honored when HASURA_GRAPHQL_DISABLE_ADMIN_SECRET=true and neither JWT nor an auth hook is configured: the server starts, and unauthenticated requests use the configured unauthorized role instead of being rejected. Admin-secret and role headers cannot override it. Setting the unauthorized role to admin while the admin secret is disabled is now a startup error. Enterprise deployments that previously combined these settings with SSO or collaborator authentication will now serve unauthenticated requests as the unauthorized role.
  • Separate outbound HTTP for tenant-defined webhooks and Remote Schemas from trusted server infrastructure, so internal engine requests are not blocked when a tenant CIDR policy is configured. This release does not enable any CIDR blocklist.

Console

  • Fix Neon integration error when creating a new data source. (Cloud only)
  • Display custom root fields in the table Modify tab in the same order as the Custom Field Names editor, with queries before mutations, unrecognized fields last, and empty values omitted.
  • Fix incorrect root-field and row-permission summaries in the Console. Omitted or null query/subscription root-field lists now display as enabled, while explicitly empty lists remain disabled and are preserved when saving permissions.
  • Improve the Run SQL page: results with duplicate column names now show each column's own value, and a "No transaction" option is added for Postgres and Citus to run statements (such as CREATE INDEX CONCURRENTLY) outside a transaction
  • Improve performance of the Browse Rows table for binary and very large values, so tables with big bytea columns no longer freeze the page. Long values are truncated in the table, and binary values are no longer rendered; a download button is shown instead.
  • Fix "Derive action" in the API explorer failing for every operation, and support aliases and inline fragments in derived actions
  • Fix action codegen crashing for actions that return a scalar type
  • Fix Python FastAPI action code generation to use the action name as the handler route and the action return type as the response model, instead of the fixed /items/ route and action arguments.
  • Fix OpenAPI-to-GraphQL path parameter substitution to preserve dollar-sign sequences literally, preventing values containing $&, $$, or similar sequences from producing incorrect request paths.
  • List Postgres table and view columns in their defined order in the console, so CSV/JSON exports and column pickers use the same order as the table or view.
  • Auto-refresh the Hasura SSO token for the GraphiQL playground.
  • Display dependent objects one per line in console error messages when dropping or untracking objects, making it easier to identify the dependencies blocking the operation.
  • Fix the console’s Run SQL “Track this” option to recognize PostgreSQL object names containing dollar signs and quoted names containing spaces, so newly created tables, views, and functions can be tracked correctly.
  • Improve the UX of adding/modifying table columns in migration mode.
  • Enhance customizations for the CSV export feature.

CLI

  • Support HASURA_GRAPHQL_LOG_LEVEL as the CLI’s default log level, allowing debug migration logs to be enabled through the process environment. An explicit --log-level flag takes precedence, and unset or blank values retain the INFO default. The variable is not read from .env files.
  • Run the binary specified by --cli-ext-path directly instead of copying it into the CLI cache, and preserve the supplied binary during cleanup. Report an error if the path does not exist or is a directory.
  • Applying failed seeds returns no-op. The error is logged, but hasura deploy --with-seeds exits with status 0. A CI user would expect a non-zero exit.
  • hasura scripts update-project-v2 no longer produces invalid SQL when merging yaml migrations into sql files
  • Fix hasura metadata export (and other commands that write metadata files) keeping the old file or directory name casing on case-insensitive file systems (macOS, Windows) when a table, function or source was renamed with only a case change, which broke the project on case-sensitive file systems. Stale files and directories whose names differ only in letter case from the exported ones are now removed or renamed.
  • Warn on hasura metadata export when a database connection string is stored as a plain-text value in databases.yaml instead of {from_env: ...}
  • Reject migration step counts of 1,000,000,000 or more passed to migrate apply --up or --down, and suggest --version <n> --type up|down for targeting a specific migration.

Data Connector

  • Updated Snowflake, MariaDB, and SQL Server JDBC drivers and Spring Framework in the V2 data connector container to address reported dependency vulnerabilities. Retained the separate JVM mitigation for CVE-2026-114938. (Enterprise/cloud only)
  • The super-connector now rejects invalid or mismatched JDBC connection strings, local filesystem connection targets, and file-loading connection options. Configurations using server-side credential, certificate, wallet or log files must migrate away from these options. (Enterprise/cloud only)

Build

  • Fix the v2 and v3 migration images applying migrations and metadata to the wrong server or skipping them when HASURA_GRAPHQL_ENDPOINT, HASURA_GRAPHQL_MIGRATIONS_DIRECTORY, HASURA_GRAPHQL_METADATA_DIRECTORY or HASURA_GRAPHQL_METADATA_FILE is set on the container.

Don't miss a new graphql-engine release

NewReleases is sending notifications on new releases.