================================================================================ Oracle Endeca Server (7.5.1.1 Revision A) * LCORE-848: Running a Dgraph on a CPU that supports the AVX instruction set (Sandy Bridge and later) may cause a Dgraph to crash. Endeca Server then issues a error similar to the following: Error contacting the conversation service on data store 'abc' at port 1234: Error parsing data version from Dgraph response. This error occurs because the OS does not support AVX. (AVX support was added in Linux kernel 2.6.30 and Windows 2008 R2 SP1.) If you have this error, contact Oracle Support. ================================================================================ Oracle Endeca Server GENERAL (7.5.1.1) - Release Date: May 2013 - This release is compatible with data indexed with Endeca Server 7.5.1.0, but not with previous releases. INSTALLATION (7.5.1.1) - See the "Oracle Endeca Server Installation Guide" for installation information. MIGRATION (7.5.1.1) - See the "Oracle Endeca Server Migration Guide" for information on migrating to this release. BUG FIXES (7.5.1.1) * LCORE-798: Improves the EQL filter performance by avoiding the bin set intersection against the bin universe during evaluation, since the intersection of a bin set against the universe always results in the same bin set. * LCORE-799: Improves the latency of RefinementGroupConfig by generating in parallel the refinements for all attributes in the group. * LCORE-770: EQL filters performed badly because the binset-extraction table function was single-threaded and could not assume its input was globally ordered. This change makes the input to the extraction function globally ordered if necessary, and that too by merging in parallel. * LCORE-812: Fix crash in text search. Add a check for an exhausted cursor in AccumulateDocumentStringIds. The check is needed because the cursor could have been used up at the end of a previous call but is still in the stringIdCursorMap. In this case, the record ID being requested is equal to the last record ID that the cursor was at in a previous invocation. Since all the work for this record ID was handled in a previous call, there is nothing to do here but check for an invalid cursor and return early. KNOWN ISSUES (7.5.1.1) * LCORE-829: The .worddat file (that is at the same level as the _indexes directory) is only a temporary file and should be deleted after an update operation (for example, after an updateSpellingDictionary() web service invocation). However, the file is not being deleted, which can cause user confusion at the Endeca Server level as this file is not preserved during a snapshot operation and also is not removed as part of a delete-dd command. Note that the file does not affect normal data domain operations. ================================================================================ Oracle Endeca Server GENERAL (7.5.1) - Release Date: March 2013 - This release is not compatible with data indexed using previous releases. - The documentation (including the release notes) for this product may have been updated since the release date. For the most recent version of the complete documentation set, see the Endeca Server documentation page on the Oracle Technology Network at: http://www.oracle.com/technetwork/middleware/endecaserver/documentation INSTALLATION (7.5.1) - See the "Oracle Endeca Server Installation Guide" for installation information. BUG FIXES (7.5.1) * LCORE-376: These changes reduce Dgraph memory usage for bulk ingest, especially for small ingests of less than 500MB of data (as transferred over the bulk ingest socket connection), which typically take 2 minutes or less. For example, for an ingest of about 500MB, the Dgraph peak RSS is reduced from about 3.5GB to 1.3GB. * LCORE-536: This fix addresses an EVE optimization problem that resulted in long running queries. * LCORE-542: This fixes an issue with the reported record counts from from bulk ingest, which in some cases could cause the client to report a negative number of records queued. This fix also includes two separate changes to the handling of the bulk ingest: 1) Messages are now buffered in a more memory-efficient format. This reduces the memory usage for both small and large ingests, with a small (roughly 2%) improvement in throughput. 2) The batch size used for splitting a bulk ingest into separate updates is now variable. It used to be fixed at 150MB; now it starts at 15MB, and grows to a maximum of 180MB after about 3000MB of data (typically about 10 minutes). This policy gives much smaller memory usage for small ingests, at the cost of slightly (1.15x) slower performance. For large ingests, performance should be roughly the same as before, possibly slightly better in some cases due to the larger maximum batch size of 180MB vs 150MB. * LCORE-545: This fixes a problem in which a precedence rule that was deleted in the Dgraph was still active until the Dgraph was re-started. Now, the deleted precedence rule is immediately deactivated from the Dgraph. KNOWN ISSUES (7.5.1) * LCORE-742: It is not possible to use fully the built-in WebLogic developer pages to access individual Endeca Server SOAP web services while running in mutual authentication SSL mode. * LCORE-752: The EndecaServer subdirectory within the DOMAIN_HOME directory is not bundled as part of the pack & unpack process for scaling out. * LCORE-766: During periods of heavy disk activity on Linux kernel versions earlier than 2.6.30, Endeca Server may experience dropped connections to the Cluster Coordinator service (ConnectionLossException), when the Cluster Coordinator is configured to store its transaction log on a busy ext3 filesystem mounted with the data=ordered journaling mode. To work around this issue, configure the Cluster Coordinator to store its transaction log on a dedicated disk or a disk without significant write activity, using the endeca-cluster-coordinator-dataLogDir option in the EndecaServer.properties file. ===============================================================================