[
https://issues.apache.org/jira/browse/HIVE-22255?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16941641#comment-16941641
]
Peter Vary commented on HIVE-22255:
-----------------------------------
[~nareshpr]: Hive full ACID is working with snapshot isolation. Here is an
example:
* We have {{table1}} and it is compacted. The directory structure is this:
{code:java}
table1/
table1/base_000001/
table1/base_000001/bucket_00000
{code}
* Query A is issued: {{select * from table1}}. The directory structure does
not change, and it will read {{table1/base_000001/bucket_00000}}
* Later Query B is issues: {{insert overwrite table1 ...}}. It will create the
following directory structure:
{code:java}
table1/
table1/base_000001/
table1/base_000001/bucket_00000
table1/base_000002/
table1/base_000002/bucket_00000
{code}
* Query B is finished. It will mark writeId 000002 as a committed
* Query C is issued: {{select * from table1}}. The directory structure does
not change, and it will read {{table1/base_000002/bucket_00000}}
* Query A is still working, an it is still reading {{base_000001}} so it can
not be removed
* Query A is finished. We can remove the {{base_000001}} directory
This means that we can have two base directories at a given time, and we are
not allowed to remove them until the last transaction reading the is not
finished.
> Hive don't trigger Major Compaction automatically if table contains only base
> files
> ------------------------------------------------------------------------------------
>
> Key: HIVE-22255
> URL: https://issues.apache.org/jira/browse/HIVE-22255
> Project: Hive
> Issue Type: Bug
> Components: Hive, Transactions
> Affects Versions: 3.1.2
> Environment: Hive-3.1.1
> Reporter: Rajkumar Singh
> Assignee: Rajkumar Singh
> Priority: Major
>
> user may run into the issue if the table consists of all base files but no
> delta, then the following condition will yield false and automatic major
> compaction will be skipped.
> [https://github.com/apache/hive/blob/master/ql/src/java/org/apache/hadoop/hive/ql/txn/compactor/Initiator.java#L313]
>
> Steps to Reproduce:
> # create Acid table
> {code:java}
> // create table myacid(id int);
> {code}
> # Run multiple insert table
> {code:java}
> // insert overwrite table myacid values(1);insert overwrite table myacid
> values(2),(3),(4){code}
> # DFS ls output
> {code:java}
> // dfs -ls -R /warehouse/tablespace/managed/hive/myacid;
> +----------------------------------------------------+
> | DFS Output |
> +----------------------------------------------------+
> | drwxrwx---+ - hive hadoop 0 2019-09-27 16:42
> /warehouse/tablespace/managed/hive/myacid/base_0000001 |
> | -rw-rw----+ 3 hive hadoop 1 2019-09-27 16:42
> /warehouse/tablespace/managed/hive/myacid/base_0000001/_orc_acid_version |
> | -rw-rw----+ 3 hive hadoop 610 2019-09-27 16:42
> /warehouse/tablespace/managed/hive/myacid/base_0000001/bucket_00000 |
> | drwxrwx---+ - hive hadoop 0 2019-09-27 16:43
> /warehouse/tablespace/managed/hive/myacid/base_0000002 |
> | -rw-rw----+ 3 hive hadoop 1 2019-09-27 16:43
> /warehouse/tablespace/managed/hive/myacid/base_0000002/_orc_acid_version |
> | -rw-rw----+ 3 hive hadoop 633 2019-09-27 16:43
> /warehouse/tablespace/managed/hive/myacid/base_0000002/bucket_00000 |
> +----------------------------------------------------+{code}
>
> you will see that Major compaction will not be trigger until you run alter
> table compact MAJOR.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
