[
https://issues.apache.org/jira/browse/YARN-7213?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16256466#comment-16256466
]
Haibo Chen edited comment on YARN-7213 at 11/17/17 4:44 AM:
------------------------------------------------------------
[~openinx], [~te...@apache.org] Here is a test that you can add to
TestTimelineReaderWebServicesHBaseStorage to reproduce the behavior
{code:java}
@Test
public void testDirectScan() throws Exception {
Table table =
util.getConnection().getTable(TableName.valueOf("prod.timelineservice.entity"));
Scan scan = new Scan();
SingleColumnValueFilter svf1 = new SingleColumnValueFilter(
Bytes.toBytes("c"), Bytes.toBytes("config_param1"),
CompareFilter.CompareOp.EQUAL, new
BinaryComparator(GenericConverter.getInstance().encodeValue("value1")));
SingleColumnValueFilter svf2 = new SingleColumnValueFilter(
Bytes.toBytes("c"), Bytes.toBytes("config_param1"),
CompareFilter.CompareOp.EQUAL, new
BinaryComparator(GenericConverter.getInstance().encodeValue("value3")));
FilterList filterList = new FilterList(FilterList.Operator.MUST_PASS_ALL);
filterList.addFilter(
new FilterList(FilterList.Operator.MUST_PASS_ONE, Arrays.asList(svf1,
svf2)));
filterList.addFilter(
new FilterList(FilterList.Operator.MUST_PASS_ONE, Arrays.asList(
new FamilyFilter(CompareFilter.CompareOp.EQUAL, new
BinaryComparator("i".getBytes()))
))
);
scan.setFilter(filterList);
Scan scan2 = new Scan(scan); // copy the scan
int i = 0;
for (Result result : table.getScanner(scan)) { i++; }
assertEquals("first scan", 2, i); // this passes
svf1.setFilterIfMissing(true);
svf2.setFilterIfMissing(true);
i = 0;
for (Result result : table.getScanner(scan2)) { i++; }
assertEquals("second scan", 2, i); // this fails
}
{code}
[~openinx] Did you apply the wip patch attached?
Copying the internal discussion from our HBase folks.
> Say you have row1:
> r1 c1 q1 v1
> r1 c1 q2 v2
> r1 c2 -- --
> With filters: SCVF(=c1/q2/v2) AND FamilyFilter(=c2)
> The filters will first get applied to r1/c1/q1/v1: the SCVF doesn't match so
> it says to skip this cell and go to the next row. The FamilyFilter >says it
> doesn't match either and it wants to go to the next family. In this
> combination, the FamilyFilter "wins" and the next cell evaluated is >r1/c2.
> Because the SCVF never saw r1/c1/q2, the filterIfMissing kicks in and then
> the whole row is dropped.
> An important thing to realize is that filters will narrow what returns, they
> are not general predicates that >determine whether the whole cell is filtered
> out or not.
> Or if this is in terms of SQL operations, HBase doesn't support something
> like SELECT x WHERE y = "foo", so you have to make do with >SELECT x,y WHERE
> y = "foo" instead.
> This was actually working before due to a bug in the filter processing that
> got fixed in 2.0
was (Author: haibochen):
[~openinx], [~te...@apache.org] Here is a test that you can add to
TestTimelineReaderWebServicesHBaseStorage to reproduce the behavior
{code:java}
@Test
public void testDirectScan() throws Exception {
Table table =
util.getConnection().getTable(TableName.valueOf("prod.timelineservice.entity"));
Scan scan = new Scan();
SingleColumnValueFilter svf1 = new SingleColumnValueFilter(
Bytes.toBytes("c"), Bytes.toBytes("config_param1"),
CompareFilter.CompareOp.EQUAL, new
BinaryComparator(GenericConverter.getInstance().encodeValue("value1")));
SingleColumnValueFilter svf2 = new SingleColumnValueFilter(
Bytes.toBytes("c"), Bytes.toBytes("config_param1"),
CompareFilter.CompareOp.EQUAL, new
BinaryComparator(GenericConverter.getInstance().encodeValue("value3")));
FilterList filterList = new FilterList(FilterList.Operator.MUST_PASS_ALL);
filterList.addFilter(
new FilterList(FilterList.Operator.MUST_PASS_ONE, Arrays.asList(svf1,
svf2)));
filterList.addFilter(
new FilterList(FilterList.Operator.MUST_PASS_ONE, Arrays.asList(
new FamilyFilter(CompareFilter.CompareOp.EQUAL, new
BinaryComparator("i".getBytes()))
))
);
scan.setFilter(filterList);
Scan scan2 = new Scan(scan); // copy the scan
int i = 0;
for (Result result : table.getScanner(scan)) { i++; }
assertEquals("first scan", 2, i); // this passes
svf1.setFilterIfMissing(true);
svf2.setFilterIfMissing(true);
i = 0;
for (Result result : table.getScanner(scan2)) { i++; }
assertEquals("second scan", 2, i); // this fails
}
{code}
[~openinx] Did you apply the wip patch attached?
Copying the internal discussion from our HBase folks.
> Say you have row1:
> r1 c1 q1 v1
> r1 c1 q2 v2
> r1 c2 -- --
> With filters: SCVF(=c1/q2/v2) AND FamilyFilter(=c2)
> The filters will first get applied to r1/c1/q1/v1: the SCVF doesn't match so
> it says to skip this cell and go to the next row. The FamilyFilter >says it
> doesn't match either and it wants to go to the next family. In this
> combination, the FamilyFilter "wins" and the next cell evaluated is >r1/c2.
> Because the SCVF never saw r1/c1/q2, the filterIfMissing kicks in and then
> the whole row is dropped.
> An important thing to realize (which I was struggling with) is that filters
> will narrow what returns, they are not general predicates that >determine
> whether the whole cell is filtered out or not.
> Or if this is in terms of SQL operations, HBase doesn't support something
> like SELECT x WHERE y = "foo", so you have to make do with >SELECT x,y WHERE
> y = "foo" instead.
> This was actually working before due to a bug in the filter processing that
> got fixed in 2.0
> [Umbrella] Test and validate HBase-2.0.x with Atsv2
> ---------------------------------------------------
>
> Key: YARN-7213
> URL: https://issues.apache.org/jira/browse/YARN-7213
> Project: Hadoop YARN
> Issue Type: Task
> Reporter: Rohith Sharma K S
> Assignee: Rohith Sharma K S
> Attachments: YARN-7213.prelim.patch, YARN-7213.wip.patch
>
>
> Hbase-2.0.x officially support hadoop-alpha compilations. And also they are
> getting ready for Hadoop-beta release so that HBase can release their
> versions compatible with Hadoop-beta. So, this JIRA is to keep track of
> HBase-2.0 integration issues.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
