We've never been able to get sqoop-merge to use compression. Even tried
setting MR output properties. I do hope you guys can figure out the issue,
would save us a lot of space.
On Tue, May 5, 2015 at 1:45 PM, Michael Arena mar...@paytronix.com wrote:
Sqoop 1.4.5-cdh5.2.1 on Cloudera CDH 5.2.1 cluster
From: Abraham Elmahrek
Reply-To: user@sqoop.apache.org
Date: Tuesday, May 5, 2015 at 4:18 PM
To: user@sqoop.apache.org
Subject: Re: Sqoop Incremental job does not honor the compression
settings after the initial run
This seems like a bug. Which version of Sqoop are you using?
On Tue, May 5, 2015 at 12:50 PM, Michael Arena mar...@paytronix.com
wrote:
I am incrementally loading data from SQL Server to Hadoop using an
Oozie Sqoop Action.
Oozie runs a saved job in the Sqoop Metastore as created below:
sqoop job \
--create import__test__mydb__mytable \
--meta-connect *** \
-- import \
--connect jdbc:sqlserver://mydbserver:1433;databaseName=mydb; \
--username \
--password-file \
--num-mappers 4 \
--target-dir /***/***/mytable \
--fields-terminated-by '\t' --input-fields-terminated-by '\t' \
--null-string '\\N' --null-non-string '\\N' \
--input-null-string '\\N' --input-null-non-string '\\N' \
--relaxed-isolation \
--query SELECT id, first_name, last_name, mod_time FROM mytable \
--split-by id \
--merge-key id \
--incremental lastmodified \
--check-column mod_time \
--last-value 1900-01-01 00:00:00.000 \
--compress --compression-codec
org.apache.hadoop.io.compress.SnappyCodec
The initial time the job runs, it creates 4 files like:
part-m-0.snappy
part-m-2.snappy
part-m-3.snappy
part-m-4.snappy
It did not need to do the merge step since there was no existing data.
However, the next time it runs, it pulls over modified rows from SQL
Server and then merges them into the existing data and creates files:
part-r-0
part-r-1
part-r-2
...
part-r-00020
part-r-00031
which are uncompressed TSV files.
The Sqoop Metastore has the compression settings saved:
% sqoop job --show import__test__mydb__mytable
...
enable.compression = true
compression.codec = org.apache.hadoop.io.compress.SnappyCodec
...
Since the files are named part-m-X.snappy after the first run, I
am guessing that the -m- in the name means the mappers created them (and
also since I specified 4 mappers).
On the second run, I am guessing that the (32?) reducers created the
output since there was merging necessary and the files have -r- in the
name.
Is this a bug or expected behavior?
Is there some other settings to tell the reducers to honor the
compression settings?
If it is a bug, where do I create an issue (JIRA) for it?
How are you engaging with millennials at your organization? Earn
“Lifetime Loyalty with Effective Millennial Engagement” by signing up for
our next webinar. Join us *Tuesday, May 12 at 1:00 EDT *to obtain the
tools you need to earn brand loyalty from this important demographic. Click
here http://content.paytronix.com/Lifetime-Loyalty_0515_sig.htmlto
register!
How are you engaging with millennials at your organization? Earn “Lifetime
Loyalty with Effective Millennial Engagement” by signing up for our next
webinar. Join us *Tuesday, May 12 at 1:00 EDT *to obtain the tools you
need to earn brand loyalty from this important demographic. Click here
http://content.paytronix.com/Lifetime-Loyalty_0515_sig.html to
register!
--
*Mauricio Aristizabal*
Manager - Business Intelligence + Data Science | Impact Radius
10 East Figueroa Street, 2nd Floor | Santa Barbara, CA 93101
m: +1 (323) 309-4260 | mauri...@impactradius.com
*Learn more – Watch our 2 minute overview
http://www.impactradius.com/?src=slsap*
www.impactradius.com | Twitter http://twitter.com/impactradius | Facebook
https://www.facebook.com/pages/Impact-Radius/153376411365183 | LinkedIn
http://www.linkedin.com/company/impact-radius-inc. | YouTube
https://www.youtube.com/user/ImpactRadius
Maximizing Return on Ad Spend