How to configure session time out time in OBIEE

Today, let's talk about something real quick and simple here. Have you ever wondered why after being idle for a while, when you resume your tasks in OBIEE, you are logged out? That's because you have wondered off for quiet sometime and the system logged you out due to inactivity. So knowing this behavior, how do we control the time that OBIEE should allow user to be idle before logging them out?

The configuration is done in instanceconfig.xml file by entering the following entries:

I have entered 60 mins here, which means the system will allow me to be idle for 60 mins before logging me out..

The place you enter these entries is shown below:

Thanks 

Until next time

Read More

a general approach to performance tuning in OBIEE

A long time ago, I wrote an article about performance tuning in OBIEE, in which I stated the importance of minimizing performance issue by having a good project direction and a clear understanding of the BI Concept. This is the ensure the project with clean design, optimized process so that we won't even running into a lot of the performance issues later on. Remember, a bad design, wrong process and mismanaged expectation are the root cause of a lot of the issues we potentially face on our path. Applying fixes, workaround and changes attempting the improve the performance are only temporary solutions that eases the symptom, but it won't help in the long run.

Although I am a strong believer of dealing with performance issue at design and project process level, it doesn't mean that there is nothing else we can do within OBIEE to optimize the performance. When a report is running very slow, there are a few areas in OBIEE that we can look into to diagnose the issue.

One of the most common place to investigate is the query log, which records the exact physical queries that a particular report is generating. When you look into the query log, don't just focus on the physical SQL, but also read the steps this log is recording. Each step of the action has a time stamp telling us what time this action started and what time it completed and moved on to the next step. Sometimes, it could be the connection to the DB that takes unusually long, not the query itself. Sometimes, we are defining some DB scripts in the connection pool for it to execute before or after queries, this can cause extra delay. So we need to look into all these things before getting to the physical query itself.

In order to tune a query, some basic understanding of our tables and objects are very important. We know that most of the queries consist of some dimension and fact tables. Fact tables are generally huge because it holds time sensitive information. All these stats, measures and keys are subject to change based on time. When you look at sales fact, you don't just look at the sales total, you need to look at sales total based on dimensional information, especially time related information in order for it to make sense. Therefore, fact tables are loaded with a certain frequency based on time, the records can be at hourly frequency or daily frequency or even min frequency. Over time, fact tables get huge. A full table scan on a fact table will definitely be very costly and slow. Therefore, when all of the queries should at least have a day dimension joining to the fact table and filter based on a specific time period to avoid full table scan. Indexes, and partitions are always needed on the fact table. But then again, if a user insists on running a report for entire 2 years of data, then there is no much we can do. He should read the myth of performance tuning.

Another thing to look for is, if the report is created with some conditional expressions to define the column values, such as case statement, make sure in the where clause the case statement is not on the right side of the equation. For example, you may have a column like x column that gives you date based on different calendars, and then create a filter (like x column = current date) out of this column. Then the where clause would look like this: Where case when column = retail then retail date when column = fiscal then fiscal date else date end = current date. This type of query will take away the index you created on fact table because it has to go through the entire table since the value is conditional.

Explain plan is always a good place to look at in order to understand what the query is actually doing. However, if the query takes way too long to run, you may not want to join the explain plan and wait for a long time. I'd suggest to analyze the query by looking at how many tables and how many joins are involved in the report. Draw a diagram if need to. Then, start removing columns and joins one after another, then rerun the report and measure the performance. I guarantee that by removing layers and layers of joins, the performance will start to get better. Then you will be able to find that 1 table, when joined with your other tables, kills the performance. Then focus on tuning a much simpler version of the query. Chances are, you need to set the driving table to be the smaller table in your logical & complex join. Understand the sequence of how data are being fetched to make sure the smaller dataset is used as the driving dataset against much larger fact tables to avoid full table scan.

An example: You have one report that fetches data from the following tables in a snowflake fashion:

Region <--> Supplier <-->Customer <--> Sales fact<-->Date

We know that sales fact is much bigger. Region has to go through supplier in order to find customer, and then go to the sales fact to find the measures for those selected customers based on date and regions. So if I go from regions to supplier and then customer, I will obtain a pretty small list of records. Then I will just search into the sales fact for these small list of customers instead of scanning the whole fact table. Other way to get the result would be, get all of the sales fact records, then take the record and search into customer and eventually get you the region. First way is much faster then the second way because the driving data set is much smaller. The explain plan will outline the flow of the joins. Analyze that and see if you are using the wrong data set as driver.

Sometimes, graphs and charts can cost system overheads too. Putting too many charts and graphs on one dashboard will slow down the performance.

Although caching, aggregate tables are available in OBIEE as features, don't look at it as magic bullets. Especially caching, it has cost and it won't work in ad-hoc environment.

Understanding your hardware resources and system environment is another important aspect of performance tuning. You need to know where your limit is.

In summary, performance tuning is a very common practice in all OBIEE project because we all face slow performance. Certain things can be avoided by having a clean design, other issues will have to be investigated and tuned at the query and DB Level, while there are still possible solutions using OBIEE features. Still there will be some issues you just can't fix. As a last resort, a deep understanding of what the customer wants is important before you apply the workaround so that you can manage the expectation properly.

Thanks

Until next time

Read More

Basic External table authentication with password encription

Hello

Today, I want to talk about some real basic stuff in OBIEE 10G, which is external table authentication. I know there are many similar posts out there on the internet, I still want to post my own version. This is not to show off or anything, just so I don't have to memorize these things in the future if the article is kept in my own space. I know that we are talking about 11G nowadays, everybody is talking about cool things they can do with 11G, google maps, unstructured data and so on. I like to get into those things too, in fact I have. But I still believe that the understanding of basic security approach in OBIEE, the usages of session variables, the standard process of user management can benefit in the long run.

So without further digression, the external table authentication is a way to manage users and user groups in an external DB table for OBIEE to authenticate user's accessibility to the application. LDAP Authentication on the other hand, is to store user information in LDAP server. For external table authentication, the definition of the table isn't always the same from company to company. But the basics are the same. You typically will have 3 tables:

First one is user table that stores user ID and other user Infos;

second one is user group table that stores user group information:

Now because 1 user can belong to many groups and 1 group can have many users, they naturally form n:n relationship. Therefore, we have a third table which map each user to user group:

Now that we have the structures ready, we will implement external table authentication in OBIEE by creating session variables. My first initialization block is 'Isauthentication' which is associated to several session variables:
User
Displayname
Loglevel
Email
Password

The follow screenshot is misleading, it should've been printed 'create 5 session variables for this block':

The initialization string for this block is:
(The bold text are the variable names with just created)



select USERNAME, DISPLAYNAME, LOGLEVEL, EMAIL,PASSWORD from IS_SEC_USER
where UPPER(USERNAME) = UPPER(':USER') and PASSWORD =  IS_GET_MD5_VAL(':PASSWORD')

This query basically goes into user table and find all of the record where column 'username' = the value of 'USER' variable and column Password = the value of the 'PASSWORD' variable.  Due to the nature of session variable, these values will be assigned when user initiates a session by attempting to log in.  The test result of this initialization block is shown:

Now what is 'IS_GET_MD5_VAL'? It is a DB function that encrypts the password in user table:

This is just an fyi, you don't have to bother about it:


create or replace FUNCTION IS_GET_MD5_VAL (p_in VARCHAR2)
   RETURN VARCHAR2
IS
   l_hash   VARCHAR2 (2000);
BEGIN
   l_hash :=RAWTOHEX(UTL_RAW.cast_to_raw(DBMS_OBFUSCATION_TOOLKIT.md5 (input_string=> p_in)));
   RETURN l_hash;
END;

So the password in the user table looks like this:

 Anyways, let's move on to the next initialization block that we will create, which in my case is authorization.
The purpose of this is to get all of the user groups that the users belong to. The previous initialization block gets the users from user table, therefore, it should take precedence over this one.

Note the row-wise initialization is needed here. It will return all of the rows that the user belong to, instead of a single value, which variable typically does.

The initialization string is this:


ISAuthorization= select 'GROUP', UG.GROUPNAME from IS_SEC_USER_USERGROUPS FFF, IS_SEC_USER U, IS_SEC_USERGROUP UG where
UG.ID = FFF.USERGROUPS and
U.ID = FFF.IS_SEC_USER and
UPPER(U.USERNAME) = UPPER(':USER')

This is clear that this query joins all of the 3 tables and find the right group that any user belong to. This is executed after the previous initialization block.

So the idea here is that when users log on, the first session will get the user's ID and password from user table if it exist, then it will join to the group table and find out what are all of the groups this user belong to. So not only user is able to log into OBIEE, but also it's privilege will also be defined based on the group privilege. We can define user/group privilege in OBIEE later.

So from previous screenshot we know there is a dummy user call 'Admin', although the password shown in the table is encrypted, but we created its password to be 'Admin'.

Now lets log on to OBIEE via Admin/Admin:

Now the result is:

Thanks

Until next time

Read More

OBIEE The application of creating sub-report to filter another report

Hello guy

In the creation of OBIEE, we all know that we can use 1 report's result set as filter condition for another report. This is called sub-report, the equivalent SQL statements from the log is then nested query. Today, I want to talk about 1 good situation where this technique can be very helpful.

Now look at the below report. This report provides a lot of PWE ID from the left.

Now, I want to create another report, the second report should only show the  PWE ID that are related to the ones from the first report when user specifies any PWE from the prompt. In other words, when user pick PWE '10083_HSPA_CD_100005', then the first report will return only for that PWE, but the second report that we are about to create should return for '10083_HSPA_CD_100006' instead of '10083_HSPA_CD_100005' because according to our rules, they are related.

So this is a perfect example of using sub-report, knowing the relation is obtained based on the prefix of the PWEs up to the last '_'. In other words, if the prefix before the last 6 digits are the same, they are related.

Subsequently, we will go back to the first report and add a new column called 'initial' with the following expression:
SUBSTRING(Pseudowire."PWE ID",1, (LENgth(Pseudowire."PWE ID")-6))

This substrings the text from '1' position, and it will go for whatever length of the text up to the last 6 digits. So in our case, the last 6 digits of each row will be chopped off:

Now let's start with the second report by adding all of the wanted fields including 'initial' with the same above expression:

Now that we have the second report which we call 'related Pseudowires', we will then start creating filters for this report. The idea is the take the column 'initial' from this report, and match it up with the same column from the first report with 'equal' operator. In this case, the second report has to return the PWE that belongs to the same initial as determined from the first report.  Then we will nest the report again by creating another filter on PWE ID. This time, PWE ID from the second report show NOT equal to the PWE ID from the first report. So when both conditions are satisfied, the only record that will come back from the second report will be the related PWE.

Now, put these reports on the dashboard. As an option, I can also define a guided navigation for section 2 where second is held. The guided navigation will act as a condition that when the second report returns at least 1 row, it will appear on the dashboard, otherwise, it's hidden.

Now, let's see how this works on the dashboard. First we don't specify the prompt value. So the first report comes back with all PWEs and second report is nowhere to be seen:

Now, select 1 PWE from the prompt and rerun the dashboard:

Now the second report appears, but it is displaying only the related PWE from the first report.

Thanks,

Until next time.

Read More

How to report data dynamically based on user's personal time zone setting -- Part 3

So from the last time I took a step further to solve the issue of reporting based on time zone, although it wasn't 100% correct, at least it opens up a few options. The conclusion there was that we need to find a way to dynamically choose the time zone offset value and use it for our time conversation. Where I was stuck was at the daylight saving changes that happens from all over the world.

Therefore, I have decided to introduce a time zone table that stores all of the time zone and the corresponding offset value. The table structure is like below, and I have manually added 2 rows, both are for GMT -6 time zone. However, one of the time zone is currently having DST in effect while the other doesn't:

Then, bring this table into OBIEE's model:

Now, let's create some dashboard prompts using these 2 columns:

Here on Time Zone prompt, I am applying the following expression to show the contend:

SELECT case when "Time Zone"."Time Zone" = '@{session.timeZone}' then "Time Zone"."Time Zone" end FROM "Capacity Management" where case when "Time Zone"."Time Zone" = '@{session.timeZone}' then "Time Zone"."Time Zone" end is not null

The previously discovered session variable 'timeZone' is truly handy here. So the idea is that, when the time zone column equals the locale name that's generated by user session, it will return that exact locale name which is also stored in the time zone table. The where clause eliminates all the other records which will be null in this case. Also, the default value of this prompt will be the session variable.

Now create the second prompt on offset value:

Now the second prompt is also coming from the same time zone table, this should record the offset value that matches the locale name in the table. To do some, in the 'SQL result' setting, I entered the following expression:

SELECT "Time Zone"."Offset Value" FROM "Time Zone" where  "Time Zone"."Time Zone" = '@{session.timeZone}'

Also, I have set a presentation variable which will pass this value into any filters or tables that need conversions. I call this variable 'Offset'.

So let's test this prompt on the dashboard with the first locale setting, which is central America time:

The result looks great. These fields are automatically populated.

Now change the locale setting to central time U.S & Canada:

Then refresh the log in again and we see it automatically updates:

Now, the rest is pretty straightforward.  Remember in my previous post where I was forced to use the following expression to convert my time stamp in my timestampadd function? 

Timestampadd(sql_tsi_day, cast(Substring(case when '@{session.timeZone}' = '(GMT) Casablanca, Monrovia' then '(GMT+00:00) Casablanca, Monrovia' when '@{session.timeZone}' = '(GMT) Greenwhich Mean Time: Dublin, Edinburgh, Lisbon, London' then '(GMT+00:00) Greenwhich Mean Time: Dublin, Edinburgh, Lisbon, London' when '@{session.timeZone}' = '' then '(GMT+00:00) Default' else '@{session.timeZone}' end From 5 for 3) as int)/24 , Date."Hour Date"

Now I can simply replace all these codes to the following:

Timestampadd(sql_tsi_day, @{offset}/24 , Date."Hour Date". 

This will take care of all scenarios..

From OBIEE's point of view, we are all done. So of course, we have to maintain our time zone table and we have to figure out what offset value is for each time zone and when is their DST start and end date of the year. I went to wikipedia and it was pretty eye opening and fun to learn all these stuffs. 

So then, I will load the entire table with all of the time zone name and offset values. Some script will have to be called on specific day to update the offset value for DST. Although this may sound complicated, it is not in real life. Think about this, most of the time we will be supporting business in Asia, Europe, North America, South America or Africa.. A lot of those time zones never observe or no longer observe DST. The main countries that do are in Europe or North America, and most of their DST period are from March to Oct or April to Nov. Then we have a few exceptions in Australia, even then, there aren't that many to deal with. Most of the time, we won't cover them all unless your BI Platform is so big that it truly need to cover users from Marshall islands, Baghdad or New Guinea, which I highly doubt. I would say, this approach takes care of most cases and should be a good solution..

Thanks, until next time 

Read More