datanucleus / datanucleus-hbase Goto Github PK

All HBase client docs on the internet are for v1.x, but it doesn't seem to work with Java 11+. See https://hbase.apache.org/book.html#java Seems that, as of Aug 2021, they still don't have full support for Java 11.

Need to decipher what needs changing.

See http://hbase.apache.org/2.2/apidocs/overview-summary.html

As a starter
HBaseAdmin -> Admin
HTableDescriptor -> TableDescriptor
HColumnDescriptor -> ColumnFamilyDescriptor

When we make use of a persistable class in the HBase plugin we currently go direct to "manageClasses", whereas we should simply check for StoreData, and only then go to HBase to check it.

We need to support where the PK is formed with a persistable object. Currently it tries to serialise the related object which is wrong

Should lookup using the converted value not the raw field value

Refer to core issue 19

Refer to datanucleus/datanucleus-core#134

When we execute a query and use FetchFieldManager, we have to use the constructor for an ExecutionContext (i.e ObjectProvider not yet known). This means that we cannot easily wrap any SCO fields in FetchFieldManager. Consequently we need to call

ObjectProvider.replaceAllLoadedSCOFieldsWithWrappers()

just after the FetchFieldManager process.

When we retrieve a List from HBase we retrieve it in the exact same order as persisted. That is, it is stored as a Collection in the owner object. If a user puts some @orderby annotation they may want the list reordering.

This applies to ALL non-RDBMS datastores, so creating some generic code would make sense and reorder them in-memory, maybe using a Comparator.

Feature Request: provide as much detail as possible about behaviour

Bug Report, attach a testcase using problem guide and state what version the problem occurs in.

So if the table is in a particular namepace other then default. It looks for a cololmn family prepending the namespace and hence throws exception with org.apache.hadoop.hbase.regionserver.NoSuchColumnFamilyException

Should use the class.isAssignableFrom(...) rather than mmd.hasCollection

See PersistenceNucleusContext.

The current HBase is 1.1.2 - but every version from 0.96 on cannot be used with DataNucleus.

refer to http://hadoop.apache.org/hbase/docs/current/api/org/apache/hadoop/hbase/client/transactional/package-summary.html and http://tephra.io/

We currently do not support embedded id classes.

When a user is utilising JDO/JPA APIs there are options to lock the objects being worked on, and then unlock them when they are updated. With RDBMS we support this via "SELECT ... FOR UPDATE" (where supported by the datastore). With HBase we could make calls to

HTable.lockRow(byte[]);
HTable.unlockRow(byte[]);

We currently build against

org.apache.hbase
hbase-client
1.1.2

Bumping this to the latest 1.x release (1.6.0) should fix it.

Trying to auto-create tables causes an error, because the table is initially created without any column family (they're added after the table creation). HBase does not allow this (anymore).
http://www.datanucleus.org/servlet/forum/viewthread_thread,7842

Refer to core issue 128 regarding database creation/deletion.

See datanucleus/datanucleus-core#81

See datanucleus/datanucleus-core#180