March 16, 2018

ODA X6-2M – Expanding Storage can be unsuccessful when impatient



In 'A Brief History of Time' (RIP Mr. Hawking), Chapter 2 is named 'Space and Time' – and this title fits perfect with a current problem, I was faced with, when expanding disk capacity in an ODA X6-2M.

Long story, told short

An ODA X6-2M, equipped with the standard disk configuration, ran out of space and my customer decided to add two NVMe's. So, he ordered two NVMe's – and, by the way - we were waiting for two weeks (!), until delivery. 

Next step - mounting the disks. According documentation, expanding disk storage is a quite easy job:
  • Put the disks in
  • Set the disks in online state (odaadmcli power disk on pd_02 and odaadmcli power disk on pd_03)
  • Expand storage (odaadmcli expand storage)

This results in my environment in a complete disaster:

  • Disks are already online when trying a 'power disk on',
  • 'expand storage' meant, the disks have already an ASM signature and aborted,
  • 'odaadmcli show disk' shows the disks, but als UNKNOWN and with the /dev/nmv* names of the existing NMVes, whilst the old ones had, all of a sudden, new names … (!)
  • The command to power off said that the disks are not online, but a subsequent remove of the disks resulted in a system crash …

In short: horror!! Opened a Service Request, tried this, tried that – and finally, after three weeks fighting with Oracle Support, I started a last try: Put the disks in, went to the coffee bar, returned about an hour later. Started to set the disks in online state (result: the disks are already online), and issued an 'odaadmcli expand storage' – expecting the same result as with every try before … but, surprisingly, the command returned no error, ASM started rebalancing and everything went fine J


Solution / Chapter 2 ... 'Space and Time'

You'll get the disk space, when You take Your time ...

Put the disks in, wait at least 15 minutes, power on the disks and issue the 'expand storage' command. Oracle Supports explaination of this behavior was (in my words): The oakd process needs some time to prepare the disks in a way that expand storage works properly …

Share:

ODA Xx – Space Waste When Using ACFS (instead of ASM) For Databases



Disclaimer: I will not discuss the necessity of an ACFS FS to store database files, nor compare ASM and ACFS or it's benefits. What to use / what fits best in Your environment is up to You, folks … ;-)

Pretty sure that this is for lots of people a well-known problem – but for all others:

When creating a database using ACFS as storage option, a FS of size 100GB is created. This 100GB is not a configurable value, by the way. Result is: When having lots of small databases, making use of ACFS wastes a lot of disk space, as every database's file system occupies 100GB.
Oracle Support's answer to that is: "In the future releases you may see the customized option to define the size of ACFS while creating the database". Nevertheless: It is possible to reduce that 100GB FS to an appropriate size. Here's how:  

Examples are based on OAK 12.2.1.2.0, used db release: 12.1.0.2

Create the database:

oracle@eoda03 ~]$ odacli create-database -m -n RCRACFS1 --no-cdb  --dbstorage ACFS -dh cce28d1b-01cc-4917-8237-38683d34f53e

… results in (from a FS perspective)

Filesystem                   Size  Used Avail Use% Mounted on
/dev/asm/datrcracfs1-262     100G  2.5G   98G   3% /u02/app/oracle/oradata/RCRACFS1

Reduce the filesystem size to 10GB:

[oracle@eoda03 ~]$ acfsutil size -90G /u02/app/oracle/oradata/RCRACFS1
acfsutil size: new file system size: 10737418240 (10240MB)

Check the filesystem size

[oracle@oda03 ~]$ df -h /u02/app/oracle/oradata/RCRACFS1
Filesystem                   Size  Used Avail Use% Mounted on
/dev/asm/datrcracfs1-262      20G  2.5G   18G  13% /u02/app/oracle/oradata/RCRACFS1

20G?? 100 minus 90 is 20? Possibly yes (you never know ;-)) – on the other hand and a better explanation: 20GB could be minimum FS for an ACFS file system.

Conclusion:

Yes! It is that easy J

Share:

March 5, 2018

EXPDP using an External Password Store - facing a new 'release' of a well known performance issue

After a while without blogging, here a new blog post talking about a very well known performance issue:

export (exp as well as expdp) is usually slower when using an TCP based TNS-Alias instead of setting ORACLE_SID. Remember? ;-) 

In a little more detail:
An "expdb system/manager@db directory= ..." takes (usually) more time than an
"export ORACLE_SID=DB; expdp system/manager directory= ..."

But what if You have to use an external password store - a wallet - to avoid clearly readable passwords either in a file or at the command line? One part of the whole procedure is, to define the TNS alias in tnsnames.ora - and most of us define a TCP based alias. This is - imho - not the best way of connecting a local database - IPC or BEQ are way better for that.

So, I solved an expdp performance issue by using an Bequeath based TNS alias:

Created a new wallet:
mkstore -wrl /u01/app/oracle/wallet -create

Created an alias to connect to the DB:
mkstore -wrl /u01/app/oracle/wallet -createCredential db_system_beq.world system manager

Added a BEQ connect description to my tnsnames.ora:
DB_SYSTEM_BEQ.WORLD =
  (DESCRIPTION =
    (ADDRESS =
          (PROTOCOL = BEQ)
          (PROGRAM = oracle)
          (ARGV0 = oracleDB)
          (ARGS = '(DESCRIPTION=(LOCAL=YES)(ADDRESS=(PROTOCOL=BEQ)))')
    )
    (CONNECT_DATA =
        (SERVICE_NAME = DB)
    )

  )

Set the environment for my db and issued an expdp:
export ORACLE_SID=DB; expdb /@db_system_beq.world directory= ...

Result:

Job "SYSTEM"."SYS_EXPORT_FULL_01" successfully completed at Mon Mar 5 15:04:15 2018 elapsed 0 00:00:43 

Result when using a TCP based TNS alias:

Job "SYSTEM"."SYS_EXPORT_FULL_01" successfully completed at Mon Mar 5 14:58:03 2018 elapsed 0 00:01:31 

Try it - and post Your results as comment.


Share:

Copyright © Robert Crames' Oracle Database and Middleware Blog | Powered by Blogger
Design by SimpleWpThemes | Blogger Theme by NewBloggerThemes.com