Mailing-List: contact commits-help@cassandra.apache.org; run by ezmlm
Precedence: bulk
Reply-To: dev@cassandra.apache.org
Date: Thu, 13 Apr 2017 08:39:41 +0000 (UTC)
From: =?utf-8?Q?Andr=C3=A9s_de_la_Pe=C3=B1a_=28JIRA=29?= <jira@apache.org>
To: commits@cassandra.apache.org
Message-ID: <JIRA.13061716.1491394637000.273917.1492072781804@Atlassian.JIRA>
In-Reply-To: <JIRA.13061716.1491394637000@Atlassian.JIRA>
References: <JIRA.13061716.1491394637000@Atlassian.JIRA> <JIRA.13061716.1491394637468@jira-lw-us.apache.org>
Subject: [jira] [Commented] (CASSANDRA-13412) Update of column with TTL
 results in secondary index not returning row
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: quoted-printable
archived-at: Thu, 13 Apr 2017 08:39:48 -0000


    [ https://issues.apache.org/jira/browse/CASSANDRA-13412?page=3Dcom.atla=
ssian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=
=3D15967286#comment-15967286 ]=20

Andr=C3=A9s de la Pe=C3=B1a commented on CASSANDRA-13412:
-----------------------------------------------

The problem is also affecting to indexes on partition key components:
{code}
CREATE TABLE k.t (
    pk1 int,
    pk2 int,
    a int,
    b int,
    PRIMARY KEY ((pk1, pk2))
);
CREATE INDEX ON k.t(pk1);

INSERT INTO k.t (pk1, pk2, a, b) VALUES (1, 2, 3, 4);
UPDATE k.t USING TTL 10 SET b =3D 10 WHERE pk1 =3D 1 AND pk2 =3D 2;
-- Wait 10 seconds

SELECT * FROM k.t WHERE pk1 =3D 1 AND pk2 =3D 2; -- 1 row
SELECT * FROM k.t WHERE pk1 =3D 1; -- 0 rows
{code}

Index entries inherit the TTL of the indexed cell, and this is right with r=
egular columns. However, indexes of primary key columns index every column,=
 meaning that their index entries will always get the TTL of the last updat=
ed row column.=20

The proposed solution is to not generate expiring cells when the indexed co=
lumn is part of the primary key. Here is the patch for 2.1 and 2.2:

||[2.1|https://github.com/apache/cassandra/compare/cassandra-2.1...adelapen=
a:13412-2.1]|[utests|http://cassci.datastax.com/view/Dev/view/adelapena/job=
/adelapena-13412-2.1-testall/]|[dtests|http://cassci.datastax.com/view/Dev/=
view/adelapena/job/adelapena-13412-2.1-dtest/]|
||[2.2|https://github.com/apache/cassandra/compare/cassandra-2.2...adelapen=
a:13412-2.2]|[utests|http://cassci.datastax.com/view/Dev/view/adelapena/job=
/adelapena-13412-2.2-testall/]|[dtests|http://cassci.datastax.com/view/Dev/=
view/adelapena/job/adelapena-13412-2.2-dtest/]|


> Update of column with TTL results in secondary index not returning row
> ----------------------------------------------------------------------
>
>                 Key: CASSANDRA-13412
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-1341=
2
>             Project: Cassandra
>          Issue Type: Bug
>            Reporter: Enrique Bautista Barahona
>            Assignee: Andr=C3=A9s de la Pe=C3=B1a
>
> Cassandra versions: 2.2.3, 3.0.11
> 1 datacenter, keyspace has RF 3. Default consistency level.
> Steps:
> 1. I create these table and index.
> {code}
> CREATE TABLE my_table (
>     a text,
>     b text,
>     c text,
>     d set<int>,
>     e float,
>     f text,
>     g int,
>     h double,
>     j set<int>,
>     k float,
>     m set<text>,
>     PRIMARY KEY (a, b, c)
> ) WITH read_repair_chance =3D 0.0
>    AND dclocal_read_repair_chance =3D 0.1
>    AND gc_grace_seconds =3D 864000
>    AND bloom_filter_fp_chance =3D 0.01
>    AND caching =3D { 'keys' : 'ALL', 'rows_per_partition' : 'NONE' }
>    AND comment =3D ''
>    AND compaction =3D { 'class' : 'org.apache.cassandra.db.compaction.Siz=
eTieredCompactionStrategy' }
>    AND compression =3D { 'sstable_compression' : 'org.apache.cassandra.io=
.compress.LZ4Compressor' }
>    AND default_time_to_live =3D 0
>    AND speculative_retry =3D '99.0PERCENTILE'
>    AND min_index_interval =3D 128
>    AND max_index_interval =3D 2048;
> CREATE INDEX my_index ON my_table (c);
> {code}
> 2. I have 9951 INSERT statements in a file and I run the following comman=
d to execute them. The INSERT statements have no TTL and no consistency lev=
el is specified.
> {code}
> cqlsh <ip> <port> -u <user> -f <file>
> {code}
> 3. I update a column filtering by the whole primary key, and setting a TT=
L. For example:
> {code}
> UPDATE my_table USING TTL 30 SET h =3D 10 WHERE a =3D 'test_a' AND b =3D =
'test_b' AND c =3D 'test_c';
> {code}
> 4. After the time specified in the TTL I run the following queries:
> {code}
> SELECT * FROM my_table WHERE a =3D 'test_a' AND b =3D 'test_b' AND c =3D =
'test_c';
> SELECT * FROM my_table WHERE c =3D 'test_c';
> {code}
> The first one returns the correct row with an empty h column (as it has e=
xpired). However, the second query (which uses the secondary index on colum=
n c) returns nothing.
> I've done the query through my app which uses the Java driver v3.0.4 and =
reads with CL local_one, from the cql shell and from DBeaver 3.8.5. All dis=
play the same behaviour. The queries are performed minutes after the writes=
 and the servers don't have a high load, so I think it's unlikely to be a c=
onsistency issue.
> I've tried to reproduce the issue in ccm and cqlsh by creating a new keys=
pace and table, and inserting just 1 row, and the bug doesn't manifest. Thi=
s leads me to think that it's an issue only present with not trivially smal=
l amounts of data, or maybe present only after Cassandra compacts or perfor=
ms whatever maintenance it needs to do.


--
This message was sent by Atlassian JIRA
(v6.3.15#6346)