5.2.0

What's Changed

• [HWORKS-3074] Importing the Python SDK executes great_expectations and takes ~9 seconds by @jimdowling in #1120
• [HWORKS-3071] Python client waits forever on a library or environment command that never reaches a final state by @jimdowling in #1118
• [FSTORE-2092] Fix feature logging and feature monitoring on feature views with model-dependent transformations by @jimdowling in #1089
• [HWORKS-2879][HWORKS-3145] Spark upgrade to 4.1. Hudi upgrade to 1.2.0 by @gibchikafa in #1042
• [HWORKS-2879][APPEND] Build the java SDK on JDK 17 in a container by @robzor92 in #1126
• [HWORKS-2879][APPEND] Give the containerized java SDK build its maven settings by @robzor92 in #1127
• [HWORKS-2879][APPEND] Give the build container a resolvable uid by @robzor92 in #1128
• [HWORKS-2879][APPEND] Bump the AWS SDK v2 in hsfs from 2.10.40 to 2.45.1 by @robzor92 in #1129
• [HWORKS-2879][APPEND] Delete Hudi records with the native delete operation by @robzor92 in #1130
• [FSTORE-2029] PushDown PIT queries to Snowflake when all feature groups in a feature view are created from same snowflake connector by @dhananjay-mk in #1115
• [HWORKS-2879][APPEND] Bump hsfs kafka-clients from 3.4.0 to 3.9.1 by @robzor92 in #1133
• [HWORKS-3074][APPEND] Silence great_expectations' metric-registry INFO chatter by @robzor92 in #1132
• [FSTORE-2101] SnowflakeConnector.connector_options() does not support key pair authentication by @dhananjay-mk in #1134
• [HWORKS-3191] hops CLI: reduce startup and login latency by @jimdowling in #1136
• [HWORKS-3185] Bump jackson and pin netty in the published hsfs uber jars by @gibchikafa in #1137
• [HWORKS-3210] hops setup: use the Python engine, print one line, and re-run cleanly by @jimdowling in #1145
• [FSTORE-2105] Don't emit KLL for a column with no finite values, and bin over the finite ones by @robzor92 in #1135
• [HWORKS-3217] Serve agent skills to every coding agent and keep their state across terminal sessions by @jimdowling in #1150
• [HWORKS-3212] Deployment schema: validated inference requests, default predictor, and feature view deployments by @jimdowling in #1147
• [FSTORE-2109] Stop training dataset writes from pinning a HybridRowQueue per shuffle partition by @robzor92 in #1155
• [FSTORE-2111] Model monitoring fan-out includes logging metadata columns and the FM job fails against a training-dataset reference by @javierdlrm in #1157
• [HWORKS-3227] Removing a member from a project leaves their home directory owned by a user with no access by @jimdowling in #1158
• [FSTORE-2082] Delete online rows from the vector database by @bubriks in #1140
• [HWORKS-2883] Add KServe Standard mode support for LLM deployments by @javierdlrm in #1122
• [HWORKS-2954] Replace livelogs pipeline with direct Kubernetes pod-log reading and HopsFS deployment log archives by @jimdowling in #1052
• [HOPSFS-406] Bump HopsFS to 3.4.3.3-EE-RC0 by @hamidgh09 in #1154
• [HWORKS-3211] hops datasource create: every data source type the platform has by @jimdowling in #1146
• Release online store connection pools instead of leaking them by @jimdowling in #1085
• [HWORKS-3209] hops CLI: readable REST errors, complete project list, search across all entity types by @jimdowling in #1142
• [HWORKS-3004] hops session CLI: push/pull/new/list/mirror/stop by @jimdowling in #1125
• [HWORKS-3234] Make spark.driver.memoryOverheadFactor and spark.executor.memoryOverheadFactor configurable for jobs, Jupyter and project defaults (UI, REST, Python client) by @manu-sj in #1169
• [HWORKS-3252] hops setup: repoint to a new cluster; add hops logout by @MagicLex in #1170
• [HWORKS-3242] Add ClickHouse as a data source by @jimdowling in #1163
• [HWORKS-2879][APPEND] Register Hudi's Spark session extension in Spark Connect sessions by @robzor92 in #1173
• [HWORKS-3260] Bump hopsworks_aiomysql to 0.2.3 by @robzor92 in #1179
• [HWORKS-3242][Append] Add ClickHouse as a data source by @jimdowling in #1174
• [HWORKS-3103] Expose the advertised vLLM image tags via REST/SDK and stop re-validating the tag on restart by @javierdlrm in #1172
• Prepare development for 5.2.0-SNAPSHOT by @robzor92 in #1192
• [HWORKS-3240] Feature view inference logging: real-time online logs from the Arrow inference logger by @jimdowling in #1164
• [HWORKS-3267] Ship an Airflow orchestration skill with the Python client by @jimdowling in #1194
• [HWORKS-2895] Add hopsworks_analytics project for global analytics with superset by @jimdowling in #1013
• [HWORKS-3240][Append] Keep the pending offsets file out of the feature group's Delta location by @jimdowling in #1198
• [HWORKS-3261] Add Teradata as a data source by @jimdowling in #1191
• [FSTORE-2116] Batch inference and training data anchored on a caller-supplied spine_df by @jimdowling in #1167
• [HWORKS-3271] Support skills for external clients by @jimdowling in #1203
• [HWORKS-3197] Stop sending a service list when creating a project by @jimdowling in #1143
• [FSTORE-2116][Append] hops td compute: refuse a time window with --split by @jimdowling in #1209
• [HWORKS-3240][Append] Materialization job: treat null write_options as none by @jimdowling in #1208
• [HWORKS-3240][Append] Encode a null timestamp in a Kafka row as null, not NaT by @jimdowling in #1210
• [HWORKS-3240][Append] Write the Kafka PEM files atomically by @jimdowling in #1211
• [HWORKS-2807][Append] Fix Hudi partitioned_by bug by @jimdowling in #1217
• [HWORKS-1408] Support separate environments for transformer and predictor by @javierdlrm in #1204
• [HWORKS-3243] Python client latency: remove redundant round trips, serialization and per-call preparation from deployment and feature paths by @jimdowling in #1165
• [HWORKS-3244] Feature-view REST client latency: cached endpoint URLs, direct projection, faster decoding and bounded waits by @jimdowling in #1166
• [HWORKS-3320] Bump hopsworks_aiomysql to 0.2.4 by @robzor92 in #1226
• [HWORKS-2881] Use python 3.13 in the base images by @robzor92 in #1227
• [FSTORE-2128] Send nmslib embedding fields the kNN query they accept in find_neighbors by @robzor92 in #1228
• [FSTORE-2107] missing init kafka offsets by @bubriks in #1199
• [HWORKS-3240][APPEND] Skip the offline insert when a materialization range has no offline records by @robzor92 in #1232
• [HWORKS-3047] strimzi kafka upgrade by @bubriks in #1213
• [FSTORE-2106] Event-time-based windows for feature monitoring configurations by @javierdlrm in #1141
• [HWORKS-2457] Support for versioning in serving deployments by @javierdlrm in #1225
• [FSTORE-1964] Allow opensearch-py 3.x by @robzor92 in #1240
• [FSTORE-2139] Retry watermark reads on a partition whose leader is not serving yet by @robzor92 in #1241
• [FSTORE-2119] insert(wait_for_online_ingestion=True) can wait on the wrong online ingestion and return early by @bubriks in #1238
• [HOPSFS-418] Bump HopsFS to 3.4.3.3-EE-RC1 by @hamidgh09 in #1242
• [HWORKS-3373] Support Python 3.14 by @robzor92 in #1245
• [HWORKS-3295] Add db_access to Python app creation by @gibchikafa in #1216
• [HWORKS-3398] Python client CI: unit tests fail with "pyarrow requires NumPy 2.0 or newer" on pyarrow 26.0.0 by @dhananjay-mk in #1255
• [HWORKS-3339] Add Elasticsearch as a data source by @jimdowling in #1234
• [HWORKS-3380] Skills: Iceberg writes from the Python engine no longer need a PYSPARK job by @robzor92 in #1249
• [HWORKS-3340] npm packages in bulk, and the Flask, Gradio and Node.js app kinds by @gibchikafa in #1246
• [HWORKS-3306] Python client: null primary key in a serving entry is handled differently by the SQL and REST clients and by single and batch reads by @dhananjay-mk in #1254
• [HWORKS-3106][HWORKS-3113] Ship KEDA: autoscale Standard-mode deployments on resource, request and vLLM engine metrics, scale to zero when idle by @gibchikafa in #1237
• [HWORKS-3387] hops-data-sources skill: use get_data_source instead of deprecated get_storage_connector by @MagicLex in #1252
• [HWORKS-3397] hops sql defaults to the project's feature groups by @MagicLex in #1253
• [HWORKS-3424] Apps: scale to zero when idle in the SDK and CLI by @gibchikafa in #1267
• [HWORKS-3275] ML software factory by @jimdowling in #1207
• [HWORKS-3370] Provided credentials for data sources (oracle) by @jimdowling in #1244
• [HWORKS-3436] start() and save() can return before the gateway routes the deployment, so the first predict gets "Deployment not created or running" by @robzor92 in #1268

Full Changelog: 5.1.0...5.2.0