[
https://issues.apache.org/jira/browse/HIVE-9445?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14297384#comment-14297384
]
Sergey Shelukhin commented on HIVE-9445:
----------------------------------------
Hmm, looks like I missed the java part of the change that was not merely code
move. Partition spec validation should not have been thrown out... Let me file
a JIRA to add it back
> Revert HIVE-5700 - enforce single date format for partition column storage
> --------------------------------------------------------------------------
>
> Key: HIVE-9445
> URL: https://issues.apache.org/jira/browse/HIVE-9445
> Project: Hive
> Issue Type: Bug
> Components: Metastore
> Affects Versions: 0.13.0, 0.14.0, 0.13.1, 0.15.0, 0.14.1
> Reporter: Brock Noland
> Assignee: Brock Noland
> Priority: Blocker
> Fix For: 0.15.0
>
> Attachments: HIVE-9445.1.patch, HIVE-9445.1.patch
>
>
> HIVE-5700 has the following issues:
> * HIVE-8730 - fails mysql upgrades
> * Does not upgrade all metadata, e.g. {{PARTITIONS.PART_NAME}} See comments
> in HIVE-5700.
> * Completely corrupts postgres, see below.
> With a postgres metastore on 0.12, I executed the following:
> {noformat}
> CREATE TABLE HIVE5700_DATE_PARTED (line string) PARTITIONED BY (ddate date);
> CREATE TABLE HIVE5700_STRING_PARTED (line string) PARTITIONED BY (ddate
> string);
> ALTER TABLE HIVE5700_DATE_PARTED ADD PARTITION (ddate='NOT_DATE');
> ALTER TABLE HIVE5700_DATE_PARTED ADD PARTITION (ddate='20150121');
> ALTER TABLE HIVE5700_DATE_PARTED ADD PARTITION (ddate='20150122');
> ALTER TABLE HIVE5700_DATE_PARTED ADD PARTITION (ddate='2015-01-23');
> ALTER TABLE HIVE5700_STRING_PARTED ADD PARTITION (ddate='NOT_DATE');
> ALTER TABLE HIVE5700_STRING_PARTED ADD PARTITION (ddate='20150121');
> ALTER TABLE HIVE5700_STRING_PARTED ADD PARTITION (ddate='20150122');
> ALTER TABLE HIVE5700_STRING_PARTED ADD PARTITION (ddate='2015-01-23');
> LOAD DATA LOCAL INPATH '/tmp/single-line-of-data' INTO TABLE
> HIVE5700_DATE_PARTED PARTITION (ddate='NOT_DATE');
> LOAD DATA LOCAL INPATH '/tmp/single-line-of-data' INTO TABLE
> HIVE5700_DATE_PARTED PARTITION (ddate='20150121');
> LOAD DATA LOCAL INPATH '/tmp/single-line-of-data' INTO TABLE
> HIVE5700_DATE_PARTED PARTITION (ddate='20150122');
> LOAD DATA LOCAL INPATH '/tmp/single-line-of-data' INTO TABLE
> HIVE5700_DATE_PARTED PARTITION (ddate='2015-01-23');
> LOAD DATA LOCAL INPATH '/tmp/single-line-of-data' INTO TABLE
> HIVE5700_STRING_PARTED PARTITION (ddate='NOT_DATE');
> LOAD DATA LOCAL INPATH '/tmp/single-line-of-data' INTO TABLE
> HIVE5700_STRING_PARTED PARTITION (ddate='20150121');
> LOAD DATA LOCAL INPATH '/tmp/single-line-of-data' INTO TABLE
> HIVE5700_STRING_PARTED PARTITION (ddate='20150122');
> LOAD DATA LOCAL INPATH '/tmp/single-line-of-data' INTO TABLE
> HIVE5700_STRING_PARTED PARTITION (ddate='2015-01-23');
> hive> show partitions HIVE5700_DATE_PARTED;
> OK
> ddate=20150121
> ddate=20150122
> ddate=2015-01-23
> ddate=NOT_DATE
> Time taken: 0.052 seconds, Fetched: 4 row(s)
> hive> show partitions HIVE5700_STRING_PARTED;
> OK
> ddate=20150121
> ddate=20150122
> ddate=2015-01-23
> ddate=NOT_DATE
> Time taken: 0.051 seconds, Fetched: 4 row(s)
> {noformat}
> I then took a dump of the database named {{postgres-pre-upgrade.sql}} and the
> data in the dump looks good:
> {noformat}
> [root@hive5700-1-1 ~]# egrep -A9 '^COPY "PARTITIONS"|^COPY
> "PARTITION_KEY_VALS"' postgres-pre-upgrade.sql
> COPY "PARTITIONS" ("PART_ID", "CREATE_TIME", "LAST_ACCESS_TIME", "PART_NAME",
> "SD_ID", "TBL_ID") FROM stdin;
> 3 1421943647 0 ddate=NOT_DATE 6 2
> 4 1421943647 0 ddate=20150121 7 2
> 5 1421943648 0 ddate=20150122 8 2
> 6 1421943664 0 ddate=NOT_DATE 9 3
> 7 1421943664 0 ddate=20150121 10 3
> 8 1421943665 0 ddate=20150122 11 3
> 9 1421943694 0 ddate=2015-01-23 12 2
> 10 1421943695 0 ddate=2015-01-23 13 3
> \.
> --
> COPY "PARTITION_KEY_VALS" ("PART_ID", "PART_KEY_VAL", "INTEGER_IDX") FROM
> stdin;
> 3 NOT_DATE 0
> 4 20150121 0
> 5 20150122 0
> 6 NOT_DATE 0
> 7 20150121 0
> 8 20150122 0
> 9 2015-01-23 0
> 10 2015-01-23 0
> \.
> {noformat}
> I then upgraded to 0.13 and subsequently upgraded the MS with the following
> command: {{schematool -dbType postgres -upgradeSchema -verbose}}
> The file {{postgres-post-upgrade.sql}} is the post-upgrade db dump. As you
> can see the data is completely corrupt.
> {noformat}
> [root@hive5700-1-1 ~]# egrep -A9 '^COPY "PARTITIONS"|^COPY
> "PARTITION_KEY_VALS"' postgres-post-upgrade.sql
> COPY "PARTITIONS" ("PART_ID", "CREATE_TIME", "LAST_ACCESS_TIME", "PART_NAME",
> "SD_ID", "TBL_ID") FROM stdin;
> 3 1421943647 0 ddate=NOT_DATE 6 2
> 4 1421943647 0 ddate=20150121 7 2
> 5 1421943648 0 ddate=20150122 8 2
> 6 1421943664 0 ddate=NOT_DATE 9 3
> 7 1421943664 0 ddate=20150121 10 3
> 8 1421943665 0 ddate=20150122 11 3
> 9 1421943694 0 ddate=2015-01-23 12 2
> 10 1421943695 0 ddate=2015-01-23 13 3
> \.
> --
> COPY "PARTITION_KEY_VALS" ("PART_ID", "PART_KEY_VAL", "INTEGER_IDX") FROM
> stdin;
> 3 2015-01-23 0
> 4 2015-01-23 0
> 5 2015-01-23 0
> 6 2015-01-23 0
> 7 2015-01-23 0
> 8 2015-01-23 0
> 9 2015-01-23 0
> 10 2015-01-23 0
> \.
> {noformat}
> Same corruption from the HIVE CLI:
> {noformat}
> hive> show partitions HIVE5700_DATE_PARTED;
> OK
> ddate=20150121
> ddate=20150122
> ddate=2015-01-23
> ddate=NOT_DATE
> Time taken: 0.844 seconds, Fetched: 4 row(s)
> hive> show partitions HIVE5700_STRING_PARTED;
> OK
> ddate=20150121
> ddate=20150122
> ddate=2015-01-23
> ddate=NOT_DATE
> Time taken: 0.114 seconds, Fetched: 4 row(s)
> hive> select * from HIVE5700_STRING_PARTED;
> OK
> some data! 2015-01-23
> some data! 2015-01-23
> some data! 2015-01-23
> some data! 2015-01-23
> Time taken: 0.536 seconds, Fetched: 4 row(s)
> hive> select * from HIVE5700_DATE_PARTED;
> OK
> some data! 2015-01-23
> some data! 2015-01-23
> some data! 2015-01-23
> some data! 2015-01-23
> Time taken: 0.145 seconds, Fetched: 4 row(s)
> {noformat}
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
