We have an external Hive table that is used for processing raw log file data. The files are hourly, and are partitioned by date and source host name.
At the moment we are importing files using simple python scripts that are triggered a few times per hour. The script creates sub folders on HDFS as needed, copies new files from the temporary local storage and adds any new partitions to Hive.
Today, new partitions are created using "ALTER TABLE ... ADD PARTITION ...". However, if another Hive query is running on the table it will be locked, which means that the add partition command will fail (if the query runs for long enough) since it requires an exclusive lock.
An alternative to this approach would be to use "MSCK REPAIR TABLE", which for some reason does not seem to aquire any locks on the table. However, I have gotten the impression that using repair table is not recommended for a production setting.
What is the best practise for adding Hive partitions programmatically in a concurrent environment?
What are the risks or disadvantages of using MSCK REPAIR TABLE?
Is there an explanation for the seemingly inconsistent locking behaviour of the two partition adding commands? I.e. do they have different effects on running queries?
Not a good answer, but we have the same issue and here are our findings :
in the Hive doc, https://cwiki.apache.org/confluence/display/Hive/Locking , locks seem pretty sensible: an 'ADD partition" will request an exclusive lock on the created partition, and a shared lock on the whole table. A SELECT query will request a shared lock on the table. So it should be fine
however, it does not work this way, at least in CDH 5.3. According to this thread, https://groups.google.com/a/cloudera.org/forum/#!topic/cdh-user/u7aM9W3pegM this is a known behavior, probably new (I am not sure, but I also think, as the author of this thread, that the issue was not there on CDH 4.7)
So basically, we're still thinking of our partition strategy, but we will probably try to create all possible partition in advance (before getting the data), as we know precisely the values of all future partitions (might not be the case for you).
Related
Which one is better (performance wise and operation on the long run) in maintaining data loaded, managed or external?
And by maintaining, i mean that these tables will have the following operations on daily basis frequently;
Select using partitions most of the time.. but for some of it they are not used.
Delete specific records, not all the partition (for example found a problem in some columns and want to delete and insert it again). - i am not sure if this supported for normal tables, unless transactional is used.
Most important, The need to merge files frequently.. may be twice a day to merge small files to gain less mappers. I know concate is available on managed and insert overwrite on external.. which one is less cost?
It depends on your use case. External table is recommended when they are used across multiple application for example Along with hive pig or other application is also used for processing the data in this kind of scenario external tables are mainly recommended.They are used when you are mainly reading data.
While in case of managed tables hive have complete control over the data. Though you can convert any external table to managed and vice versa
alter table table_name SET TBLPROPERTIES('EXTERNAL'='TRUE');
As in your case you are doing frequent modifications in data so it is better that hive should have total control over the data. In this scenraio it is recommended to use Managed tables.
Apart from that managed table are more secure then external table because external table can be accessed by anyone. While in managed table you can implement hive level security which provided better control but in case of external you will have to implement HDFS level security.
You can refer the below links which can give you few pointers in considerations
External Vs Managed tables comparison
It's kinda real-world problem and I believe the solution exists but couldn't find one.
So We, have a Database called Transactions that contains tables such as Positions, Securities, Bogies, Accounts, Commodities and so on being updated continuously every second whenever a new transaction happens. For the time being, We have replicated master database Transaction to a new database with name TRN on which we do all the querying and updating stuff.
We want a sort of monitoring system ( like htop process viewer in Linux) for Database that dynamically lists updated rows in tables of the database at any time.
TL;DR Is there any way to get a continuous updating list of rows in any table in the database?
Currently we are working on Sybase & Oracle DBMS on Linux (Ubuntu) platform but we would like to receive generic answers that concern most of the platform as well as DBMS's(including MySQL) and any tools, utilities or scripts that can do so that It can help us in future to easily migrate to other platforms and or DBMS as well.
To list updated rows, you conceptually need either of the two things:
The updating statement's effect on the table.
A previous version of the table to compare with.
How you get them and in what form is completely up to you.
The 1st option allows you to list updates with statement granularity while the 2nd is more suitable for time-based granularity.
Some options from the top of my head:
Write to a temporary table
Add a field with transaction id/timestamp
Make clones of the table regularly
AFAICS, Oracle doesn't have built-in facilities to get the affected rows, only their count.
Not a lot of details in the question so not sure how much of this will be of use ...
'Sybase' is mentioned but nothing is said about which Sybase RDBMS product (ASE? SQLAnywhere? IQ? Advantage?)
by 'replicated master database transaction' I'm assuming this means the primary database is being replicated (as opposed to the database called 'master' in a Sybase ASE instance)
no mention is made of what products/tools are being used to 'replicate' the transactions to the 'new database' named 'TRN'
So, assuming part of your environment includes Sybase(SAP) ASE ...
MDA tables can be used to capture counters of DML operations (eg, insert/update/delete) over a given time period
MDA tables can capture some SQL text, though the volume/quality could be in doubt if a) MDA is not configured properly and/or b) the DML operations are wrapped up in prepared statements, stored procs and triggers
auditing could be enabled to capture some commands but again, volume/quality could be in doubt based on how the DML commands are executed
also keep in mind that there's a performance hit for using MDA tables and/or auditing, with the level of performance degradation based on individual config settings and the volume of DML activity
Assuming you're using the Sybase(SAP) Replication Server product, those replicated transactions sent through repserver likely have all the info you need to know which tables/rows are being affected; so you have a couple options:
route a copy of the transactions to another database where you can capture the transactions in whatever format you need [you'll need to design the database and/or any customized repserver function strings]
consider using the Sybase(SAP) Real Time Data Streaming product (yeah, additional li$ence is required) which is specifically designed for scenarios like yours, ie, pull transactions off the repserver queues and format for use in downstream systems (eg, tibco/mqs, custom apps)
I'm not aware of any 'generic' products that work, out of the box, as per your (limited) requirements. You're likely looking at some different solutions and/or customized code to cover your particular situation.
I am new to Teradata & fortunately got a chance to work on both DDL-DML statements.
One thing I observed is Teradata is very slow when time comes to UPDATE the data in a table having large number of records.
The simplest way I found on the Google to perform this update is to write an INSERT-SELECT statement with a CASE on column holding values to be update with new values.
But what when this situation arrives in Data Warehouse environment, when we need to update multiple columns from a table holding millions of rows ?
Which would be the best approach to follow ?
INSERT-SELECT only OR MERGE-UPDATE OR MLOAD ?
Not sure if any of the above approach is not used for this UPDATE operation.
Thank you in advance!
At enterprise level, we expect volumes to be huge and updates are often part of some scheduled jobs/scripts.
With huge volume of data, Updates comes as a costly operation that involve risk of blocking table for some time in case the update fails (due to fallback journal). Although scripts are tested well, and failures seldom happen in production environments, it's always better to have data that needs to be updated loaded to a temporary table in required form and inserted back to same table after deleting matching records to maintain SCD-1 (Where we don't maintain history).
We have data that is partitioned by date. Our pipeline everyday add new partition to table. We also have use cases, where people are running long read queries by filtering on partitions.
We have seen the behavior, where, If there is some read query running on old partitions (Having SHARED lock), then we can't acquire lock for new partition (EXCLUSIVE).
Is there a way that HIVE recognize that, new partition will not be read by earlier query and acquire EXLUSIVE lock for that partition to increase parallelism.
According to the documentation having a shared lock on a partition, does not prevent you from getting an exclusive lock on that same partition.
In theory your question would then appear invalid, however in practice I could think of the following two things:
It is too hard for hive to see which partitions you are using (complicated/indirect where statement?)
The documentation is not accurate, or misinterpreted by me
I would suggest you to check for point 1, and otherwise add a minimal reproducible example.
Advice/suggestions needed for a bit of application design.
I have an application which uses 2 tables, one is a staging table, which many separate processes write to, once a 'group' of processes has finished, another job comes along a aggregates the results together into a final table, then deletes that 'group' from the staging table.
The problem that I'm having is that when the staging table is being cleared out, lots of redo is generated and I'm seeing a lot of 'log file sync' waits in the database. This is a shared database with many other applications and this is causing some issues.
When applying the aggregate, the rows are reduced to about 1 row in the final table for every 20 rows in the staging table.
I'm thinking of getting around this by rather than having a single 'staging' table, I will create a table for each 'group'. Once done, this table can just be dropped, which should result in much less redo.
I only have SE, so partitioned tables isn't an option. Also faster disks for the redo probably isn't an option in the short term either.
Is this a bad idea? Any better solutions to be offered?
Thanks.
Would it be possible to solve the problem by having your process do a logical delete (i.e. set a DELETE_FLAG column in the table to 'Y') and then having a nightly process that truncates the table (potentially writing any non-deleted rows to a separate table before the truncate and then copy them back after the table is truncated)?
Are you certain that the source of the log file sync waits is that your disks can't keep up with the I/O? That's certainly possible, of course, but there are other possible causes of excessive log file sync waits including excessive commits. There is an excellent article on tuning log file sync events on the Pythian blog.
The most common cause of excessive log file syncs is too frequent commits, which are often deliberately coded in a mistaken attempt to reduce system load due to locking. You should commit only when your business transaction is complete.
Loading each group into a separate table sounds like a fine plan to reduce redo. You can truncate individual group table following each aggregation.
Another (but I think probably worse) option is to create a new staging table with the groups that haven't been aggregated then drop the original and rename the new table to replace the staging table.
I prefer Justin's suggestion ("logical delete"), but another option to consider might be a partitioned table, if you have the EE licence. The aggregation process could drop a partition instead of deleting the rows.