[HBASE-6912] Filters are not properly applied in certain cases

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 0.94.1
• Fix Version/s: 0.94.2
• Assignee: Lars Hofhansl
• Reporter: Alex Newman
• Created: 01/Oct/12 23:40
• Updated: 07/Apr/13 05:07
• Resolved: 04/Oct/12 00:41

Description

Steps to reproduce:

Create a table, load data into it. Flush the table.

Do a scan with
1. Some filter which should not match the first entry in the scan
2. Where one specifies a family and column.
You will notice that the first entry is returned even though it doesn't match the filter.

It looks like the when the first KeyValue of a scan in the column from the point of view of the code

HRegion.java

} else if (kv != null && !kv.isInternal() && filterRowKey(currentRow)) {

Is generated by

public static KeyValue createLastOnRow(final byte [] row,
final int roffset, final int rlength, final byte [] family,
final int foffset, final int flength, final byte [] qualifier,
final int qoffset, final int qlength) { return new KeyValue(row, roffset, rlength, family, foffset, flength, qualifier, qoffset, qlength, HConstants.OLDEST_TIMESTAMP, Type.Minimum, null, 0, 0); }

So it is always internal from that point of the code.

Only later from within
StoreScanner.java

public synchronized boolean next(List<KeyValue> outResult, int limit, String metric) throws IOException {
....
LOOP: while((kv = this.heap.peek()) != null) {

( The second time through)

Do we get the actual kv, with a proper type and timestamp. This seems to mess with filtering.

Issue Links

• relates to: HBASE-6562 Fake KVs are sometimes passed to filters