DBA Blogs

Partition Storage -- 1 : Default Partition Sizes in 12c

Hemant K Chitale - Tue, 2016-04-19 10:17
11g 11.2.0.2 introduced a change whereby the default Initial Extent of a Table Partition was 8MB.  However, this did not apply to Index Partitions which could still start with 64KB extents in an AutoAllocate Tablespace.

12cR1 now introduces a parameter to enable large Initial Extent for Index Partitions as well.

SQL> connect / as sysdba                          
Connected.
SQL> select banner from v$version where banner like 'Oracle Database%';

BANNER
--------------------------------------------------------------------------------
Oracle Database 12c Enterprise Edition Release 12.1.0.2.0 - 64bit Production

SQL> show parameter deferred

NAME TYPE VALUE
------------------------------------ ----------- ------------------------------
deferred_segment_creation boolean TRUE
SQL>
SQL> connect hemant/hemant
Connected.
SQL> create table my_part_tbl (id_column number(6), data_column varchar2(100))
2 partition by range(id_column)
3 (partition p_100 values less than (101),
4 partition p_200 values less than (201),
5 partition p_300 values less than (301),
6 partition p_400 values less than (401),
7 partition p_max values less than (maxvalue))
8 /

Table created.

SQL> create index my_part_tbl_ndx on my_part_tbl(id_column) local;

Index created.

SQL>
SQL> select segment_name, partition_name, segment_type, bytes/1024
2 from user_segments
3 where segment_name like 'MY_PART_%'
4 order by 1,2;

no rows selected

SQL> 
SQL> insert into my_part_tbl values (51,'Fifty One');

1 row created.

SQL> insert into my_part_tbl values (151,'One Hundred Fifty One');

1 row created.

SQL> insert into my_part_tbl values (251, 'Two Hundred Fifty One');

1 row created.

SQL> select segment_name, partition_name, segment_type, bytes/1024
2 from user_segments
3 where segment_name like 'MY_PART_%'
4 order by 1,2;

SEGMENT_NAME PARTITION_NA SEGMENT_TYPE BYTES/1024
------------------------------ ------------ ------------------ ----------
MY_PART_TBL P_100 TABLE PARTITION 8192
MY_PART_TBL P_200 TABLE PARTITION 8192
MY_PART_TBL P_300 TABLE PARTITION 8192
MY_PART_TBL_NDX P_100 INDEX PARTITION 64
MY_PART_TBL_NDX P_200 INDEX PARTITION 64
MY_PART_TBL_NDX P_300 INDEX PARTITION 64

6 rows selected.

SQL>


I can enable large Index Partition Extent with a parameter (which can be set with ALTER SESSION)


SQL> alter session set "_index_partition_large_extents"=TRUE;

Session altered.

SQL> insert into my_part_tbl values (351,'Three Hundred Fifty One');

1 row created.

SQL> select segment_name, partition_name, segment_type, bytes/1024
2 from user_segments
3 where segment_name like 'MY_PART_%'
4 order by 1,2;

SEGMENT_NAME PARTITION_NA SEGMENT_TYPE BYTES/1024
------------------------------ ------------ ------------------ ----------
MY_PART_TBL P_100 TABLE PARTITION 8192
MY_PART_TBL P_200 TABLE PARTITION 8192
MY_PART_TBL P_300 TABLE PARTITION 8192
MY_PART_TBL P_400 TABLE PARTITION 8192
MY_PART_TBL_NDX P_100 INDEX PARTITION 64
MY_PART_TBL_NDX P_200 INDEX PARTITION 64
MY_PART_TBL_NDX P_300 INDEX PARTITION 64
MY_PART_TBL_NDX P_400 INDEX PARTITION 8192

8 rows selected.

SQL>


However, I can rebuild the Index Partition Extent as well :

SQL> alter index my_part_tbl_ndx rebuild partition p_400 storage (initial 64K);

Index altered.

SQL> select segment_name, partition_name, segment_type, bytes/1024
2 from user_segments
3 where segment_name like 'MY_PART_%'
4 order by 1,2;

SEGMENT_NAME PARTITION_NA SEGMENT_TYPE BYTES/1024
------------------------------ ------------ ------------------ ----------
MY_PART_TBL P_100 TABLE PARTITION 8192
MY_PART_TBL P_200 TABLE PARTITION 8192
MY_PART_TBL P_300 TABLE PARTITION 8192
MY_PART_TBL P_400 TABLE PARTITION 8192
MY_PART_TBL_NDX P_100 INDEX PARTITION 64
MY_PART_TBL_NDX P_200 INDEX PARTITION 64
MY_PART_TBL_NDX P_300 INDEX PARTITION 64
MY_PART_TBL_NDX P_400 INDEX PARTITION 64

8 rows selected.

SQL>


In the next post, we'll see more Extents for the Partitions.
,
,
,
Categories: DBA Blogs

SQL Server Dates, Dates and More Dates

Pythian Group - Tue, 2016-04-19 09:44

 

Working with SQL Server date functions can be frustrating. This purpose of this blog is to share some date statements I use regularly, especially when doing business Intelligence and DataWarehouse solutions.

I hope you find them useful and if you have any questions or any more useful statements in relation to dates in SQL Server, please feel free to leave them in the comments below

----Today
SELECT GETDATE() 'Today'
----Yesterday
SELECT DATEADD(d , -1 , GETDATE()) 'Yesterday'
----First Day of Current Week
SELECT DATEADD(wk , DATEDIFF(wk , 0 , GETDATE()) , 0) 'First Day of Current Week'
----Last Day of Current Week
SELECT DATEADD(wk , DATEDIFF(wk , 0 , GETDATE()) , 6) 'Last Day of Current Week'
----First Day of Last Week
SELECT DATEADD(wk , DATEDIFF(wk , 7 , GETDATE()) , 0) 'First Day of Last Week'
----Last Day of Last Week
SELECT DATEADD(wk , DATEDIFF(wk , 7 , GETDATE()) , 6) 'Last Day of Last Week'
----First Day of Current Month
SELECT DATEADD(mm , DATEDIFF(mm , 0 , GETDATE()) , 0) 'First Day of Current Month'
----Last Day of Current Month
SELECT DATEADD(ms , -3 , DATEADD(mm , 0 , DATEADD(mm , DATEDIFF(mm , 0 , GETDATE()) + 1 , 0))) 'Last Day of Current Month'
----First Day of Last Month
SELECT DATEADD(mm , -1 , DATEADD(mm , DATEDIFF(mm , 0 , GETDATE()) , 0)) 'First Day of Last Month'
----Last Day of Last Month
SELECT DATEADD(ms , -3 , DATEADD(mm , 0 , DATEADD(mm , DATEDIFF(mm , 0 , GETDATE()) , 0))) 'Last Day of Last Month'
----First Day of Current Year
SELECT DATEADD(yy , DATEDIFF(yy , 0 , GETDATE()) , 0) 'First Day of Current Year'
----Last Day of Current Year
SELECT DATEADD(ms , -3 , DATEADD(yy , 0 , DATEADD(yy , DATEDIFF(yy , 0 , GETDATE()) + 1 , 0))) 'Last Day of Current Year'
----First Day of Last Year
SELECT DATEADD(yy , -1 , DATEADD(yy , DATEDIFF(yy , 0 , GETDATE()) , 0)) 'First Day of Last Year'
----Last Day of Last Year
SELECT DATEADD(ms , -3 , DATEADD(yy , 0 , DATEADD(yy , DATEDIFF(yy , 0 , GETDATE()) , 0))) 'Last Day of Last Year'
Categories: DBA Blogs

Internals of Querying the Concurrent Requests’ Queue – Revisited for R12.2

Pythian Group - Tue, 2016-04-19 09:01

Once upon a time I wrote about the Internal Workflow of an E-Business Suite Concurrent Manager Process. Many things have changed since that blog post, the most obvious change being the release of Oracle e-Business Suite R12.2. I decided to check if the way the concurrent manager queues were processed by concurrent manager processes were still the same. My main goal was to see if the manager processes still don’t attempt any way of coordination to distribute the requests among them.

This is how I did the testing:

  • I used the VM templates provided by Oracle to build my R12.2.4 test environment. By the way, I didn’t expect that the process of getting the environment up would be so simple! Downloading the media files from edelivery.oracle.com was the most time-consuming step, once done – it took me just 1 hour to un-compress everything, import the Virtual Assembly file and bring up the R12.2.4 environment on my laptop.
  • 3 Standard managers are defined by default
  • Sleep seconds were left as is = 30 seconds
  • Cache size was increased from 1 to 5.
  • Identified the 3 DB processes that belong to the Standard managers:
    select sid, serial# from v$session where module='e:FND:cp:STANDARD'
  • I enabled tracing with binds and waits for each of them like this:
    exec dbms_monitor.session_trace_enable(sid,serial#,true,true);
  • Once that was done I submitted one concurrent program – “Active users” and waited for it to complete.
  • I disabled the tracing and collected the trace files.
    exec dbms_monitor.session_trace_disable(sid,serial#);
  • Collected the trace files

I found 2 of the trace files to be very interesting. To make things more simple, the manager process “A” will be the one that executed the concurrent request, and process “B” will be the one that didn’t.

Before the “Active Users” Request Was Submitted

No other requests were running at the time I did the testing, so I clearly observed how both Managers A and B queried the FND_CONCURRENT_REQUESTS table BOTH of the trace files displayed the same method of how requests are picked up from the queue. Note, I’m showing only the lines relevant to display the main query only, and I have formatted the query text to make it more readable:

PARSING IN CURSOR #139643743645920 len=1149 dep=0 uid=100 oct=3 lid=100 tim=1460211399835915 hv=3722997734 ad='d275f750' sqlid='cd23u4zfyhvz6'
SELECT R.Rowid
FROM Fnd_Concurrent_Requests R
WHERE R.Hold_Flag                             = 'N'
AND R.Status_Code                             = 'I'
AND R.Requested_Start_Date                   <= Sysdate
AND (R.Node_Name1                            IS NULL
OR (R.Node_Name1                             IS NOT NULL
AND FND_DCP.target_node_mgr_chk(R.request_id) = 1))
AND (R.Edition_Name                          IS NULL
OR R.Edition_Name                            <= sys_context('userenv', 'current_edition_name'))
AND EXISTS
  (SELECT NULL
  FROM Fnd_Concurrent_Programs P
  WHERE P.Enabled_Flag         = 'Y'
  AND R.Program_Application_Id = P.Application_Id
  AND R.Concurrent_Program_Id  = P.Concurrent_Program_Id
  AND EXISTS
    (SELECT NULL
    FROM Fnd_Oracle_Userid O
    WHERE R.Oracle_Id = O.Oracle_Id
    AND EXISTS
      (SELECT NULL
      FROM Fnd_Conflicts_Domain C
      WHERE P.Run_Alone_Flag = C.RunAlone_Flag
      AND R.CD_Id            = C.CD_Id
      )
    )
  AND (P.Execution_Method_Code                          != 'S'
  OR (R.PROGRAM_APPLICATION_ID,R.CONCURRENT_PROGRAM_ID) IN ((0,98),(0,100),(0,31721),(0,31722),(0,31757)))
  )
AND ((R.PROGRAM_APPLICATION_ID,R.CONCURRENT_PROGRAM_ID) NOT IN ((510,40032),(510,40033),(510,42156),(510,42157),(530,43793),(530,43794),(535,42626),(535,42627),(535,42628)))
ORDER BY NVL(R.priority, 999999999),
  R.Priority_Request_ID,
  R.Request_ID
END OF STMT
EXEC #139643743645920:c=0,e=33,p=0,cr=0,cu=0,mis=0,r=0,dep=0,og=1,plh=3984653669,tim=1460211399835910
FETCH #139643743645920:c=0,e=546,p=0,cr=106,cu=0,mis=0,r=0,dep=0,og=1,plh=3984653669,tim=1460211399836507
WAIT #139643743645920: nam='SQL*Net message to client' ela= 3 driver id=1952673792 #bytes=1 p3=0 obj#=-1 tim=1460211399836572

*** 2016-04-09 10:17:09.837
WAIT #139643743645920: nam='SQL*Net message from client' ela= 30000367 driver id=1952673792 #bytes=1 p3=0 obj#=-1 tim=1460211429836965
...
EXEC #139643743645920:c=0,e=59,p=0,cr=0,cu=0,mis=0,r=0,dep=0,og=1,plh=3984653669,tim=1460211429838767
FETCH #139643743645920:c=0,e=689,p=0,cr=106,cu=0,mis=0,r=0,dep=0,og=1,plh=3984653669,tim=1460211429839587
WAIT #139643743645920: nam='SQL*Net message to client' ela= 4 driver id=1952673792 #bytes=1 p3=0 obj#=-1 tim=1460211429839652

*** 2016-04-09 10:17:39.840
WAIT #139643743645920: nam='SQL*Net message from client' ela= 30000325 driver id=1952673792 #bytes=1 p3=0 obj#=-1 tim=1460211459840003
...

It’s important to observe that:

  • All manager’s processes still compete for the same requests. If the query is executed at the same time, the same list of concurrent requests will be retrieved by all processes.
  • The constants literals used in lines 30-32 mean that the query for checking the queue is still built when the concurrent manager process starts up. These constants are mainly used to implement the specializations rules in the query.
  • Only rowid for the pending requests’ rows in FND_CONCURRENT_REQUESTS are fetched.
  • The sleep time is clearly visible on lines 41,42 and 48,49
After the “Active Users” Request Was Submitted – Starting the Concurrent Request

The manager process A was the first to pick up the submitted requests and it could be observed by the “r=1” (1 row fetched) in the FETCH call for the query we just reviewed:

FETCH #139643743645920:c=0,e=437,p=0,cr=113,cu=0,mis=0,r=1,dep=0,og=1,plh=3984653669,tim=1460211519844640

Immediately after this, the manager process A locked the row in FND_CONCURRENT_REQUESTS table, this way, the request got assigned to this process. Notice the similar where predicates used in this query, these are actually required to make sure that the request is still not picked up by another manager process. However the main thing here is the fact that the request row is accessed by the “rowid” retrieved earlier (row 45, the value of the bind variable “:reqname” is “AAAjnSAA/AAAyn1AAH” in this case). Locking of the row is done by the “FOR UPDATE OF R.status_code NoWait” clause on line 49:

PARSING IN CURSOR #139643743640368 len=4530 dep=0 uid=100 oct=3 lid=100 tim=1460211519864113 hv=4239777398 ad='cde86338' sqlid='6ya6bzgybbrmq'
SELECT R.Conc_Login_Id,
  R.Request_Id,
  ... excluded other 156 columns for brevity...
FROM fnd_concurrent_requests R,
  fnd_concurrent_programs P,
  fnd_application A,
  fnd_user U,
  fnd_oracle_userid O,
  fnd_conflicts_domain C,
  fnd_concurrent_queues Q,
  fnd_application A2,
  fnd_executables E,
  fnd_conc_request_arguments X
WHERE R.Status_code             = 'I'
AND (R.Edition_Name            IS NULL
OR R.Edition_Name              <= sys_context('userenv', 'current_edition_name'))
AND R.Request_ID                = X.Request_ID(+)
AND R.Program_Application_Id    = P.Application_Id(+)
AND R.Concurrent_Program_Id     = P.Concurrent_Program_Id(+)
AND R.Program_Application_Id    = A.Application_Id(+)
AND P.Executable_Application_Id = E.Application_Id(+)
AND P.Executable_Id             = E.Executable_Id(+)
AND P.Executable_Application_Id = A2.Application_Id(+)
AND R.Requested_By              = U.User_Id(+)
AND R.Cd_Id                     = C.Cd_Id(+)
AND R.Oracle_Id                 = O.Oracle_Id(+)
AND Q.Application_Id            = :q_applid
AND Q.Concurrent_Queue_Id       = :queue_id
AND (P.Enabled_Flag            IS NULL
OR P.Enabled_Flag               = 'Y')
AND R.Hold_Flag                 = 'N'
AND R.Requested_Start_Date     <= Sysdate
AND ( R.Enforce_Seriality_Flag  = 'N'
OR ( C.RunAlone_Flag            = P.Run_Alone_Flag
AND (P.Run_Alone_Flag           = 'N'
OR NOT EXISTS
  (SELECT NULL
  FROM Fnd_Concurrent_Requests Sr
  WHERE Sr.Status_Code         IN ('R', 'T')
  AND Sr.Enforce_Seriality_Flag = 'Y'
  AND Sr.CD_id                  = C.CD_Id
  ))))
AND Q.Running_Processes                                     <= Q.Max_Processes
AND R.Rowid                                                  = :reqname
AND ((P.Execution_Method_Code                               != 'S'
OR (R.PROGRAM_APPLICATION_ID,R.CONCURRENT_PROGRAM_ID)       IN ((0,98),(0,100),(0,31721),(0,31722),(0,31757))))
AND ((R.PROGRAM_APPLICATION_ID,R.CONCURRENT_PROGRAM_ID) NOT IN ((510,40032),(510,40033),(510,42156),(510,42157),(530,43793),(530,43794),(535,42626),(535,42627),(535,42628))) 
FOR UPDATE OF R.status_code NoWait

The behavior of the manager process B was a little bit more interesting. It too managed to fetch the same rowid from FND_CONCURRENT_PROCESSES table belonging to the submitted “Active Users” processes. However, when it tried to lock the row in FND_CONCURRENT_REQUESTS (By using exactly the same query), this happened:

PARSING IN CURSOR #139690311998256 len=4530 dep=0 uid=100 oct=3 lid=100 tim=1460211519900924 hv=4239777398 ad='cde86338' sqlid='6ya6bzgybbrmq'
...
BINDS #139690311998256:
...
Bind#2
  oacdty=01 mxl=32(18) mxlc=00 mal=00 scl=00 pre=00
  oacflg=20 fl2=1000001 frm=01 csi=873 siz=0 off=64
  kxsbbbfp=7f0c2f713f20  bln=32  avl=18  flg=01
  value="AAAjnSAA/AAAyn1AAH"
EXEC #139690311998256:c=1000,e=1525,p=0,cr=25,cu=1,mis=0,r=0,dep=0,og=1,plh=4044729389,tim=1460211519902727
ERROR #139690311998256:err=54 tim=1460211519902750

The query failed with “ORA-00054: resource busy and acquire with NOWAIT specified or timeout expired”.
This is how the access to pending concurrent requests is serialized to make sure only one of the manager processes can run it. And, I think, relying on the well-tuned and highly efficient locking mechanism of Oracle Database is a very very smart idea.

Conclusions
  • The coordination between manager processes is still not happening to distribute the requests, but the managers all query the queue the same way and then compete between themselves to lock the requests’ entries on the table 1st. The process that gets the lock also gets to execute the concurrent request.
  • The cache size variable couldn’t be observed in the trace files, but as far as I remember from my previous research the process would only fetch “cache size”-number of rowids using the 1st query in this post. This could be tested by submitting larger volume of requests simultaneously.
  • The “sleep seconds” kicks in only when the manager process didn’t fetch any rowids from the queue. After all the cached requests are attempted/executed by the manager process, the queue is checked again immediately without waiting for the “sleep seconds” (Not explained in detail in this post, but it’s revealed in the trace files)
  • The DMLs used to query the FND_CONCURRENT_REQUESTS and to lock the row are very very similar to Pre-R12.2 releases of e-Business Suite (Another sign that the process hasn’t changed, though one change that I see is the addition of where clause predicates for Checking the Editions).
Categories: DBA Blogs

Links for 2016-04-18 [del.icio.us]

Categories: DBA Blogs

Partner Webcast – Collaboration in a digital world with Oracle Documents Cloud

The modern digital workplace requires a new set of capabilities, one that leverages the best in consumerization, social engagement and business empowerment. The key to this transformation are the...

We share our skills to maximize your revenue!
Categories: DBA Blogs

More Effective Anti-Entropy Repair in Cassandra

Pythian Group - Mon, 2016-04-18 13:12
1. Introduction

Cassandra offers three different repair mechanisms to make sure data from different replicas are consistent: Hinted Hand-off, Read-Repair, and Anti-Entropy Repair.

The first two mechanisms are kind of “automatic” mechanisms that will be triggered internally within Cassandra, depending on the configuration parameter values that are set for them either in cassandra.yaml file (for Hinted Hand-off) or in table definition (for Read-Repair). The last mechanism (Anti-Entropy Repair), however, needs to be triggered with manual intervention, by running “nodetool repair” command (possibly with various options associated with it).

Despite the “manual” nature of Anti-Entropy repair, it is nevertheless necessary, because the first two repair mechanisms cannot guarantee fixing all data consistency scenarios for Cassandra. For example, 1) what if a node is down longer than “max_hint_window_in_ms” (which defaults to 3 hours)? and  2) For deletes, Anti-Entropy repair has to be executed before “gc_grace_seconds” (defaults to 10 days) in order to avoid tombstone resurrection.

At the same time, however, due to how Anti-Entropy repair is designed and implemented (Merkle-Tree building and massive data streaming across nodes), it is also a very expensive operation and can cause a big burden on all hardware resources (CPU, memory, hard drive, and network) within the cluster. In this case, how to run Anti-Entropy repair effectively becomes a constant topic within Cassandra community.

In this post, I’d like to explore and summarize some of the techniques that can help achieve a more effective Anti-Entropy repair for Cassandra.

2. What is the issue of an Anti-Entropy Repair

Before Cassandra version 2.2, a sequential full Anti-Entropy repair is the default behavior when “nodetool repair” command (with no specific options) is executed. When this happens,

1) The node that initiates the operation, called coordinator node, scans the partition ranges that it owns (either as primary or replica) one by on. For each partition range, it sends the request to each of the peer/replica nodes to build a Merkle tree.

2) The peer/replica node scans all SSTables and a major, or validation, compaction is triggered, which reads every row in the SSTables, generates a hash for it, and then adds the result to a Merkle tree.

3) Once the peer node finishes building the Merkle tree, it sends the result back to the coordinator. The coordinator node compares every Merkle tree with all other trees. If difference is detected, data is exchanged between differing nodes.

Looking at this highly summarized procedure, there are a few things that immediately caught my attention:

First, because building Merkletree requires hashing every row of all SSTables,  it is a very expensive operation, stressing CPU, memory, and disk I/O.

Second, when Merkletree difference is detected, network bandwidth can be overwhelmed to stream the large amount of data to remote nodes.

Lastly, although not obvious, it is a worse problem in my opinion, which is that this operation can cause computation repetition and therefore waste resources unnecessarily. Please see this post for an example of how this can happen.

Based on the high cost related with it, a default (pre-Cassandra 2.2) sequential full Anti-Entropy repair is, in practice, rarely considered to be a routine task to run in production environment. Actually, a common trick that many people do with Cassandra repair is simply touching all data and let read-repair do the rest of work.

However, there do have situations when an Anti-Entropy repair is required, for example, to recover from data loss. What can we do in these cases?

3. What can help to achieve a more effective Anti-Entropy repair?

Over the time, different options for “nodetool repair” command have been introduced in different Cassandra versions. These options represent different techniques that can help achieve more effective Anti-Entropy repair, which I’ll go through in more details in this section. Meanwhile, I will also try to clarify some of the confusion that new Cassandra users might have around the different nodetool repair command options (and related jargon’s) of different Cassandra versions.

3.1 Primary range repair

In Cassandra, each node manages several (token) ranges of data. One of them is the primary range which is the token range that is assigned to the node according to the token setup within the ring. Other ranges of data are actually replica of primary ranges from other nodes. Running “nodetool repair” with option “-pr” (or “–partition-range”)  on a node means that the node only repairs the data of the primary range, but not other ranges managed on this node. The default behavior of a repair (without this option) is to repair all ranges of data managed by a node.

When using this option, it avoids the cost of doing Merkle tree calculation on non-primary range data. It also helps reduce excessive data streaming across the network. One caveat of using this option is that since each node only repairs one range of data that is managed by a node, this option needs to run on ALL nodes in the ring in order to repair all the data.

This option is available in very early release (0.x) of Cassandra.

3.2 Sequential repair

“nodetool repair” option “-snapshot” (or “–with-snapshot“) means a sequential repair (also called snapshot repair) and is a feature introduced in Cassandra version 1.2 and made as default in Cassandra version 2.0. Actually, in DataStax document for Cassandra 2.0 and later, you won’t find this option. In Cassandra 2.0 and beyond, you can specify option “-par” (or “–parallel”) to tell Cassandra to run in parallel. What does this really mean?

As we mentioned in section 3.1, each node manages several different ranges of data, either as primary or replica. When “-par” option is used, it instructs Cassandra to run repair on all these ranges at the same, which means the expensive repair operation (especially the Merkletree building part) happens on multiple nodes concurrently. This could be problematic and may slow down the entire cluster.

But when using “-snapshot” option (or default in Cassandra 2.0 and beyond), for each range of data, a snapshot is first taken and the repair operation is executed on the snapshot sequentially. This means that at any time for a given replica set of data, only one replica is being repaired, allowing other replica to satisfy the application requests in a more performant way.

3.3 Incremental repair

Since Cassandra version 2.1, a new option “-ic” (or “–incremental “) is introduced for incremental repair. Starting from Cassandra version 2.2, this option becomes the default option and at the same time, option “-full” (or “–full”) is used to specify a full repair.

The option of incremental repair is a great feature that will help overcome all 3 issues as listed in Section 2. Instead of building a Merkle tree out of all SSTables (repaired or not), this option only builds the tree out of un-repaired SSTables. Therefore,

  • The size of the Merkle tree to be built is way smaller and therefore requires much less computing resources
  • When doing Merkle tree comparison, less data will be compared and potentially streamed over the network
  • Already repaired data don’t need to be computed and compared again, thus avoiding a lot of unnecessary repetition.

There are a few things that need to pay attention to when using this option, please check my previous post  Apache Cassandra 2.1 Incremental Repair for more details.

Please also be noted that when running incremental repair in Cassandra 2.1 with Leveled Compaction Strategy (LCS), it may fail with RuntimeException (see CASSANDRA-9935 for more detail).

3.4 Sub-range repair

So far in this post when talking about repairing a range of data, it means the entire range (either primary or non-primary), which is sometimes also called as endpoint range. Beginning with Cassandra version 1.1.11, “nodetool repair” has options of “-st” (or “–start-token”) and “-et” (or “–end-token”) to specify a particular sub-range of data to repair.

Conceptually, sub-range repair is much like primary range repair, except that each sub-range repair operation focuses even smaller subset of data. So in general sub-range repair shares much of the pros and cons as primary range repair. One key benefit of using sub-range repair is the freedom of specifying the repair granularity through command line, which gives the DevOps team much more flexibility regarding how to set up a repair schedule to best match the cluster’s workload status. It is also doable to consider parallel running of multiple sub-range repairs on different sets of data at the same time, either on the same node, or on different nodes. This option has actually been deemed as one of the advanced repair techniques, as per post: Advanced repair techniques

Despite the flexibility of this technique, it has to be emphasized that no matter how the repair schedule is set for sub-range repairs, all ranges of data in the ring has to be repaired at least once before gc_grace_seconds limit is reached.

3.4.1 How to calculate valid sub-range token values

When invalid sub-range values are provided to the “-st” or “-et” option of “nodetool repair” command, most likely the command will fail and throw errors about “invalid tokens”. So in order to make sub-range repair work effectively, we need a systematic method that can generate valid sub-range token values.

We know that in Cassandra, data spans multiple nodes using a token ring. Each node is responsible for one or more slices of that ring. Each slice is a token range that has the start and end point. By this understanding, a natural way to generate valid sub-range token values would be: 1) find out all token ranges associated with one node in the entire ring; 2) for each token range associated with the node, divide the range into smaller chunks of sub-ranges. The start and end point of these sub-ranges would be valid values to be fed into the “-st” or “-et” option of “nodetool repair” command. The actual method and granularity to divide a node’s token range into smaller sub-ranges is where the flexibility comes from and can be adjusted accordingly to best suit the needs of the system.

There are different ways to find token ranges that are associated with a Cassandra node. Some of them are summarized below:

  1. Run “nodetool ring” command. “nodetool status” command will also do for single-token set-up.
  2. Run CQL query against “local” or “peers” tables in “system” keyspace to get the token values associated with a host.
  3. Cassandra client drivers provides APIs to get such information as well. For example, the “Metadata” class (in package com.datastax.driver.core) of the Java client API  provides the following two methods:
    * public Set<TokenRange> getTokenRanges() :
      Returns the token ranges that define data distribution in the ring.
    * public Set<TokenRange> getTokenRanges(String keyspace,Host host) :
      Returns the token ranges that are replicated on the given host, for the given keyspace.
  4. Based on the thrift Java API describe_splits call, an open-source utility tool called “cassandra-list-subranges” has been developed to list valid sub-range values for a specific keyspace and table on a particular Cassandra node. For details of this tool, please check the GitHub repository page at https://github.com/pauloricardomg/cassandra-list-subranges

Sub-range repair is available since Cassandra version 1.1.11.

3.4.2 DataStax Enterprise (DSE) repair service

If the Cassandra cluster is running under DataStax Enterprise version, the OpsCenter provides a “repair service” starting from version 4.0. The way that this service works is to continuously repairing small chunks of data (using sub-range repair) in a cluster in the background until the entire cluster is repaired and then it starts the next cycle of processing.

From the OpsCenter Window, there is only one parameter to specify the maximum amount of time it takes to repair the entire cluster once, which is typically less than gc_grace_seconds setting. The default is 9 days, compared with default 10 days of gc_grace_seconds setting.

More advanced setting, such as the parallelism of concurrent repairs, the maximum pending repairs allowed to be running on a node at one time, and etc. can be set either in opscenterd.conf file (for all clusters) or in cluster_name.conf file (for a specific cluster).

More detailed information about DSE repair service and its configuration can be found in the following documents:

Please be noted that DSE repair service, although convenient, is only available for DSE version and not for open source Cassandra version. It also lacks the capability to specify what keyspaces or tables  to repair, such as in “nodetool repair” tool.

3.5 Data-Center vs. Cluster wide repair

Since Cassandra version 1.2, Cassandra starts to provide options for “nodetool repair” to offer the following repair capabilities regarding repair locality/scope:

  • Cassandra 1.2: “-local” (or “–in-local-dc”), only repair nodes in the same data center as the node on which the “nodetool repair” command is executed.
  • Cassandra 2.0: “-dc <dc_name>” (or “–in-dc <dc_name>”), repairs nodes in the named data center
  • Cassandra 2.1:
    • “-dcpar” (or “–dc-parallel”), repairs data center in parallel;
    • “-hosts” (or “–in-hosts”), repairs specified hosts

When none of these options is specified, all data replica across the entire cluster is repaired.

4. Conclusion

In this post, we discussed in details some of the techniques that can help with a more effective Anti-Entropy repair in Cassandra. To summarize:

  • Primary range repair is recommended for maintenance purpose and when it is used, it has to be executed on all nodes in the ring one by one to make sure the whole data range in the ring is covered.
  • When available (Cassandra version 2.1 and beyond), incremental repair will bring the biggest benefit compared with other techniques. But please be aware of the caveats related with it. Please also be noted that with the bug as specified in CASSANDRA-9935, incremental repair may have issues with Level Compaction Strategy (LCS).
  • Sub-range repair can be very useful, especially when incremental repair is not an option. But please make sure that whatever schedule that is set up using this technique has to make sure all range of data to be finished within gc_grace_seconds limit.
  • If a repair is needed to recover from data loss, a sequential, full repair is needed. Neither primary range repair nor incremental repair technique works properly for this purpose. But due to the high cost associated with this operation, sometimes it might be faster to simply wipe out the data directory on a node and let it do bootstrapping.
Categories: DBA Blogs

9 Days of Learning About Engineered Advantages

APRIL 2016 ...

We share our skills to maximize your revenue!
Categories: DBA Blogs

java.lang nullpointerexception While Installing 12.1.0.2 RAC Database Software

Oracle in Action - Mon, 2016-04-18 04:41

RSS content

I have an Oracle  12.1.0.2c cluster having following 5 nodes :

  • Hub Nodes : host01, host02, host03
  • Leaf nodes: host04 and host05

I wanted to install RAC database software on the 3 hub nodes i.e. host01, host02 and host03.

I invoked the OUI as oracle user (Owner of Database Home)

[oracle@host01 database_12_1_0_2]$ ./runInstaller

and chose the option to install RAC database software only on the 3 hub nodes.

After all the pre-requisite checks were successful , I clicked the Install button.
I got the error: java.lang nullpointerexception

On clicking OK, the OUI aborted.

To troubleshoot, I ran the OUI in debug mode as :

[oracle@host01 database_12_1_0_2]$ ./runInstaller -debug -J-DTRACING.ENABLED=true -J-DTRACING.LEVEL=2 -J-DSRVM_TRACE_LEVEL=2 -J-DFULLTRACE

The trace file showed that the leaf node host05 was not responding:

[setup.flowWorker] [ 2016-04-18 14:33:22.771 IST ]
[CRSCTLUtil.getClusterNodeConfigRoles:1198] outputs[4] = CRS-4404: The following nodes did not reply within the allotted time:
[setup.flowWorker] [ 2016-04-18 14:33:22.772 IST ]
[CRSCTLUtil.getClusterNodeConfigRoles:1200] values length = 2
[setup.flowWorker] [ 2016-04-18 14:33:22.772 IST ]
[CRSCTLUtil.getClusterNodeConfigRoles:1198] outputs[5] = host05
[setup.flowWorker] [ 2016-04-18 14:33:22.772 IST ]
[CRSCTLUtil.getClusterNodeConfigRoles:1200] values length = 1

I realized that although I wanted to install database software on 3 nodes only, all the nodes
needed to be up and running whereas in my case, the node host05 was down at that time.

When I started the node host05 and reinvoked OUI, the database software was installed
successfully.

References:
http://www.hhutzler.de/blog/oracle-installer-and-rac-installation/



Tags:  

Del.icio.us
Digg

Copyright © ORACLE IN ACTION [java.lang nullpointerexception While Installing 12.1.0.2 RAC Database Software], All Right Reserved. 2016.

The post java.lang nullpointerexception While Installing 12.1.0.2 RAC Database Software appeared first on ORACLE IN ACTION.

Categories: DBA Blogs

Online Relocation of Database File : ASM to FileSystem and FileSystem to ASM

Hemant K Chitale - Sun, 2016-04-17 10:44
There have been few published examples of the online datafile relocation feature in 12c.  The examples I've seen are on filesystem.

Here I show online relocation to/from ASM and FileSystem.

SQL> connect system/oracle
Connected.
SQL> create tablespace test_relocate;

Tablespace created.

SQL> create table test_relocate_tbl
2 tablespace test_relocate
3 as select * from dba_objects;

Table created.

SQL> select tablespace_name, bytes/1024
2 from user_segments
3 where segment_name = 'TEST_RELOCATE_TBL';

TABLESPACE_NAME BYTES/1024
------------------------------ ----------
TEST_RELOCATE 13312

SQL> select file_name, bytes/1024
2 from dba_data_files
3 where tablespace_name = 'TEST_RELOCATE';

FILE_NAME
--------------------------------------------------------------------------------
BYTES/1024
----------
+DATA/NONCDB/DATAFILE/test_relocate.260.909444793
102400


SQL>
SQL> alter database move datafile
2 '+DATA/NONCDB/DATAFILE/test_relocate.260.909444793'
3 to '/oradata/NONCDB/test_relocate_01.dbf';

Database altered.

SQL> !ls -l /oradata/NONCDB
total 102408
-rw-r----- 1 oracle asmdba 104865792 Apr 17 23:39 test_relocate_01.dbf

SQL> 
SQL> alter database move datafile  
2 '/oradata/NONCDB/test_relocate_01.dbf'
3 to '+DATA';

Database altered.

SQL> select file_name, bytes/1024
2 from dba_data_files
3 where tablespace_name = 'TEST_RELOCATE';

FILE_NAME
--------------------------------------------------------------------------------
BYTES/1024
----------
+DATA/NONCDB/DATAFILE/test_relocate.260.909445261
102400


SQL>
SQL> !ls -l /oradata/NONCDB
total 0

SQL>


Note that I was courageous enough to not use the KEEP keyword (which is optional !).
.
.
.

Categories: DBA Blogs

Links for 2016-04-16 [del.icio.us]

Categories: DBA Blogs

What’s Holding You Back From Giving Back?

Pythian Group - Fri, 2016-04-15 10:19

This week in honour of National Volunteer Week, I’m reflecting on the importance that volunteering has had in my life.

I’ve learned from each and every one of my experiences. From wrapping holiday presents at the mall, to helping source articles for an industry magazine, to wish granting, recruiting committee and board members, and providing HR advice and counsel. These experiences eventually led me to become a member of the board of directors for a number of organizations, and I was even named Board Chair.

Ironically, the rewards and benefits that I have received from the experiences far outweigh any amount of time I have given the many organizations I have supported over the years. Volunteering has provided me the opportunity to expand my skills and experience, and take on leadership roles long before I had the credentials to be hired for them. I initially started volunteering when I moved to Ottawa, and quickly learned that there is no better way to get to know your community, meet new people and expand your network. Once I started, I never looked back. I caught the “volunteer bug.” It is an important part of my life.

I am often asked how I find the time to volunteer. I always respond with, “like anything, if it’s important to you, you can and will find the time.” As I have expanded my family and career, I seek opportunities where I can continue to share my knowledge, skills and experience in ways that do not impede on either. A perfect example of this would be career mentoring. I have been a mentor for a number of organizations including the HRPA, OCISO, and the WCT. I have been fortunate to have great mentors in the past and now pay it forward. I remain connected with many of them.

In my role as VP of HR at Pythian I was thrilled to champion our Love Your Community Programs. These programs provide our employees in over 36 countries with a volunteer day and opportunities for sponsorship – i.e. raising money for causes that are meaningful to them. The programs have allowed Pythian the opportunity to positively impact the communities where our employees live.

Volunteer Week highlights the importance of volunteering in our communities, and showcases the impact it has on the lives of both the volunteer, and the communities they support. What’s holding you back from giving back?

And because it couldn’t be said any better: “We make a living by what we get, but we make a life by what we give.”

Winston Churchill

Categories: DBA Blogs

Optimizer Stew – Parsing the Ingredients to Control Execution Plans

Pythian Group - Fri, 2016-04-15 09:01

No matter how many times I have worked with Outlines, Baselines and Profiles, I keep having to look up reminders as to the differences between these three.

There is seemingly no end to articles to the number of articles and blog that tell you what needs to be licensed, how to use them, and which version of Oracle where each made its debut appearance.

This blog will discuss none of that.  This brief article simply shows the definitions of each from the Glossary for the most current version of the Oracle databases. As of this writing that version is 12.1.0.2.

And here they are.

Stored Outline

A stored outline is simply a set of hints for a SQL statement. The hints in stored outlines direct the optimizer to choose a specific plan for the statement.

Link to Stored Outline in the Oracle Glossary

SQL plan baseline

A SQL baselines is a set of one or more accepted plans for a repeatable SQL statement. Each accepted plan contains a set of hints, a plan hash value, and other plan-related information. SQL plan management uses SQL plan baselines to record and evaluate the execution plans of SQL statements over time.

Link to SQL Plan Baseline in the Oracle Glossary

SQL profile

A SQL profile is a set of auxiliary information built during automatic tuning of a SQL statement. A SQL profile is to a SQL statement what statistics are to a table. The optimizer can use SQL profiles to improve cardinality and selectivity estimates, which in turn leads the optimizer to select better plans.

Link to SQL Profile in the Oracle Glossary

Categories: DBA Blogs

Partner Webcast – Transition to the New Integration Model with Oracle SOA Cloud Service

Do you want to fully integrate your enterprise, using the same integration tool and skills for both cloud and on premises deployment? Oracle’s hybrid integration platform allows you to extract...

We share our skills to maximize your revenue!
Categories: DBA Blogs

Log Buffer #469: A Carnival of the Vanities for DBAs

Pythian Group - Wed, 2016-04-13 09:40

This Log Buffer Edition digs deep into the realms of Oracle, SQL Server and MySQL and brings together a few of the top blog posts.

Oracle

We’ve all encountered a situation when you want to check a simple query or syntax for your SQL and don’t have a database around. Of course, most of us have at least a virtual machine for that, but it takes time to fire it up, and if you work from battery, it can leave you without power pretty quickly.

View Criteria is set to execute in Database mode by default. There is option to change execution mode to Both. This would execute query and fetch results from database and from memory.  Such query execution is useful, when we want to include newly created (but not committed yet) row into View Criteria result. Newly created row will be included into View Criteria resultset.

Upgrading database hardware in an organization is always a cumbersome process. The most time consuming step is, planing for the upgrade, which mainly includes choosing right hardware for your Oracle databases. After deciding on the hardware type for your databases, rest will be taken care by technical teams involved.

Gluent New World #02: SQL-on-Hadoop with Mark Rittman

The pre-12c implementation of DCD used TNS packages to “ping” the client and relied on the underlying TCP stack which sometimes may take longer. Now in 12c this has changed and DCD probes are implemented by TCP Stack. The DCD probes will now use the TCP KEEPALIVE socket.

 

SQL Server

Snippets will allow you to code faster by inserting chunks of code with few key strokes.

One of more common concerns among database administrators who consider migrating their estate to Azure SQL Database is their ability to efficiently manage the migrated workloads.

A SQL Server Patching Shortcut

Move an Existing Log Shipping Database to a New Monitor Server

Knee-Jerk Performance Tuning : Incorrect Use of Temporary Tables

 

MySQL

MySQL 5.7 sysbench OLTP read-only results: is MySQL 5.7 really faster?

7 Galera Cluster presentations in Percona Live Santa Clara 18-21.4. Meet us there!

Generate JSON Data with dbForge Data Generator for MySQL v1.6!

Extending the SYS schema to show metadata locks

EXPLAIN FORMAT=JSON wrap-up

Categories: DBA Blogs

Links for 2016-04-12 [del.icio.us]

Categories: DBA Blogs

Installing SQL Server 2016 – Standalone Instance – New Features

Pythian Group - Tue, 2016-04-12 10:48

In this article I am going to go through a typical install of SQL Server 2016, and explain some of the best practices to follow when setting up a production server. I will also take a look at the new features when installing SQL Server 2016 compared to older versions.

The version of SQL Server I am using in SQL Server 2016 RC2. That means that some of the features may change between the time of writing and the retail release of SQL Server 2016.

 

Let’s Get Started!

The first screen you come to after opening the installation media:

Initial Screen

  1. Click the Installation button highlighted, and then the first option in the list New SQL Server Standalone Installation.

 

New for SQL Server 2016
For the Seasoned SQL Server DBA you will notice a few additions to this screen.

  • SQL Server Management Tools can now be installed from this screen. The files and binaries will be downloaded when you click this link, as they are no longer bundled with the installation media.
  • SQL Server Data Tools can also be installed from this screen.
  • Additionally, a standalone instance of R can be installed. R is a statistical programming language embedded into SQL server 2016, making it easy for data scientists and BI professionals to get a good level of analysis without leaving the SQL Server environment.

 

  1. Select the edition of SQL SERVER you would like you would like to install, I chose developer as it’s a full featured installation of SQL Server that can be used for development only, and not in a production environment. You could also enter a key in here instead of selecting a version.

Edition

  1. Read and accept the licence.
  2. SQL Server will then check a few rules to make sure you can install SQL server on your hardware.
  3. Select whether or not to allow Microsoft to check for updates and click next.
  4. SQL Server Installation will then install some setup files needed for installation and perform a few more checks. As you can see below, a warning has appeared asking me to make sure the correct ports are open on the firewall. Click Ok and then Next to proceed with the installation.

Rule Check

  1. The next screen is where we want to select all the features we want to install. As my machine is being set up as a dev machine to test SQL 2016 features, I am installing all the features. In a production environment only install the features that are needed. You can also select where to install the binaries and the root instance from this screen. Click next once you have selected the settings needed.

Features

New for SQL Server 2016

  • The ability to install a standalone instance of R is now available in the Shared Features Section.
  • The ability to install R services in the database engine is now available.
  • The ability to install Polybase Query Service for external sources is now available to install. Polybase query service allows users to query big data sources such as Hadoop using common T-SQL statements. If you are planning on installing the Polybase feature, then the Java Runtime SRE needs to be installed first.

 

  1. The next screen is where you will need to name the instance, if this is not the only SQL server installation on this hardware. It also confirms the installation directory as per the previous screens. As this is the only installation of SQL Server on this machine, I am going to leave these settings as default. Click next when ready to proceed.

Default Instance

  1. The Next Screen is the Server Configuration screen. You should run each service under a domain account with a strong password. Whether you use managed accounts or do not enforce password expiration is up to you. However, these accounts are going to be running your services and should always be available. As this is a test machine not connected to a domain, I will leave the defaults. You can also select your default server collation from this window by clicking on the tab at the top highlighted in yellow. It is important also to set the start-up type parameters to allow services to start automatically on reboot if needed.

The ability to allow Perform Volume Maintenance tasks to be checked from this screen is a new feature. This is a best practice among SQL server DBA to allow instant file initialization. Previously this had to be done outside of the installation window.

Server Config

 

New for SQL Server 2016

  • The ability to allow Perform Volume Maintenance tasks to be checked from the Server Configuration Screen.

 

  1. Next up is the database configuration screen, and we are going to step through the tables.

In the first tab you will want to add all the users that require sysadmin access. To administer the server, click on the Add Current user and Add buttons. I always use mixed mode authentication so I can still get to the server if Active Directory plays up, and I add a complex password for security. When your done, click on the Data Directories tab at the top.

Database Config1

This is the screen where you set up all of your default directories for your databases.

Best practice states that we should put Log files on separate disks to Data Files, as there are two different access patterns for these, so they would be more performant on separate disks. Backups should also be on their own disks where possible. Additionally, the OS should also have its own drive separate from all SQL server files. As this is a test server and I only have a c drive, I will leave them as default. Click TempDB tab when ready.

Database Config 2

 

New For SQL 2016

  • This is new in SQL 2016, and previously had to be configured after install. This screen allows us to create files for temp db. Best practice stated there should be 1 file per logical core up to a maximum of 8 as I have 4 cores in my machine I have created 4 files. You can also spread the files over more than one disk if needed. Once you’re happy with your selections click next.

Database Config 3

 

  1. The next screen is to configure Analysis Services. I have configured mine in Multidimensional mode adding myself as sysadmin and setting directories using same best practice as database engine. Look out for a further blog article on SQL server Analysis Services.

SSAS

 

  1. Leave the default options for Reporting Services. Again keep an eye out for another article on Reporting Services.
  2. In the next screen you have to configure the users capable of using the Distributed Replay Controller.
  3. Give a name to the DRC.

DRC_Name

  1. On the next Screen you will need to accept the terms of the features being installed by clicking accept, and then next.
  2. Finally, you can click install and that’s it! SQL Server 2016 and all its new features are installed.

Final

 

Here is a great link on some of the new features available in SQL Server 2016

Categories: DBA Blogs

SQL Server 2016 : A New Security Feature – Always Encrypted

Pythian Group - Tue, 2016-04-12 09:15

Security. This word is so important when it comes to data, and there is a reason why. Every business has it’s vital data, and this data has to be accessed only by those who are authorized. Back in 2009, I wrote an article on what measures to take when it comes to securing SQL Server.

There were days when we used to have a third party tool to encrypt the data inside SQL Server. Later, Microsoft introduced Transparent Data Encryption (TDE) bundled with the release of SQL Server 2008. You may be wondering why it is so important to encrypt the data. Inside our database, there may be a case that the customer/application has to enter and store the sensitive information such as Social Security Number (SSN) or Financial/Payment Data, which should not be read in plain text, even by a DBA. With this requirement, a strong encryption and/or data masking comes into the picture.

With the launch of SQL Server 2016 Release Candidate 0 (RC0) , Microsoft has introduced two new features that are my personal favorite – 1)  Always Encrypted and 2) Dynamic Data Masking. Today I am going to walk you through the Always Encrypted feature.

First and foremost, we need to have SQL Server 2016 RC0 installed so that we can test this feature. You can download the RC0 here.  Once you are ready with RC0, create a test database and a table with an Encrypted column to store the sensitive data. There are few prerequisites that I will list for you here. If you want, you can use the sample schema from MS.

  1. Create a sample database
  2. Create Column Master Key
  3. Generate a self-signed certificate (well, you will need to install this certificate on the machine where the application will run)
  4. Configure Column Encryption Key
  5. Create a test table with Always Encrypted column
  6. Create an application to Insert data into the sample table we created in previous step

I have created a sample app and a demo script for the reference which you can download here. Basically, what we have to remember is that we can not insert the value inside the Always Encrypted table directly, we will need to use the tool/app, and the data will always be encrypted when it goes inside the database. This will ensure that the intruder can not get the data as it travels to the database in a cipher text form.

Here is some further reading on this topic. Enjoy reading and testing an excellent feature of SQL Server 2016 RC0.

Categories: DBA Blogs

SQL On The Edge #9 – Azure SQL Database Threat Detection

Pythian Group - Mon, 2016-04-11 15:30

Despite being well documented for several years now, every now and then we still run into clients that have bad experiences because of SQL injection attacks. If you’re not familiar, a SQL injection attack happens when an attacker exploits an application vulnerability in how they pass queries and data into the database and insert their own malicious SQL code to be executed. If you want to see different examples and get the full details, the Wikipedia page is very comprehensive.

Depending on how the application is configured, this kind of attack can go all the way from enabling attackers to see data they shouldn’t, to dropping an entire database if your application is allowed to do so. The fact that it’s an application based vulnerability also means that it really depends on proper coding and testing of all inputs in the application to prevent it. In other words, it can be very time-consuming to go back and plug all the holes if the application wasn’t securely built from the ground up.

Built-in Threat Detection

To attack this issue, and as part of the ongoing security story of SQL Server, Microsoft has now invested in the feature called Database Threat Detection. When enabled, the service will automatically scan the audit records generated from the database and will flag any anomalies that it detects. There are many patterns of injections so it makes sense to have a machine be the one reading all the SQL and flagging them. MS is not disclosing the patterns or the algorithms in an effort to make working around the detection more difficult.

What about on-premises?

This feature right now is only available on Azure SQL Db. However, we all know that Azure SQL Db is basically the testing grounds for all major new features coming to the box product. I would not be surprised if the threat detection eventually makes it to the on-premises product as well (my speculation though, nothing announced about this).

Pre-requisites
For this new feature you will need Azure SQL Db, you will also need to have auditing enabled on the database. The current way this works is by analyzing the audit records so it’s 100% reactive, nothing proactive. You will need a storage account as well since that’s where the audit logs get stored. The portal will walk you through this whole process, we’ll see that in the demo video.

Current State
As I mentioned, right now the tool is more of a reactive tool as it only lets you know after it has detected the anomaly. In the future, I would love to see a preventive configuration where one can specify a policy to completely prevent suspicious SQL from running. Sure, there can always be false alarms, however, if all the application query patterns are known, this number should be very low. If the database is open to ad-hoc querying then a policy could allow to only prevent the queries or even shut down the database after several different alerts have been generated. The more flexible the configuration, the better, but in the end what I want to see is a move from alerting me to preventing the injection to begin with.

In the demo, I’m going to go through enabling Azure SQL threat detection, some basic injection patterns and what the alerts look like. Let’s check it out!

 

Categories: DBA Blogs

Enter the Exadata X6

Pythian Group - Mon, 2016-04-11 15:20

Data is exploding and Exadata is catching up. With the proliferation of cloud technology and in-memory databases; Oracle Exadata X6-2 and X6-8 has it all. It seems to be an ideal platform for hyper-convergence for any data center running Oracle products.

Following are some of the salient features of Oracle X6:

  • The compute nodes have twenty two-core Intel Xeon E5-2699 v4 processors
  • The memory is DDR4 and of size 256Gb and it can be expanded to 768Gb.
  • The local storage can now be upgraded to 8 drives from default of 4.
  • The cell servers have ten-core Intel Xeon E5-2630 v4 processors.
  • Flashcache has become massive here reaching up to 12.8TB. Full rack has 179TB of flash.
  • There will be up to 1.7 PB of disk capacity (raw) per rack.
  • The infiniband network will have 40Gb/second. There is no change to Infiniband. In X5 it became active/active – but that’s the only difference.
  • On the software side, there are many improvement but one thing which caught my eye is the feature that enables Storage Indexes to be moved along the data when a disk hits predictive failure or true failure. This surely will improve performance by a long way.

With X6, Exadata has surely come a long way forward.

Categories: DBA Blogs

Pages

Subscribe to Oracle FAQ aggregator - DBA Blogs