input file
output file
lookup file
intermediate file

2.Database Components

3.Partition Components

4.De-Partition Components

5.Sort Components

sort
sort within groups
partition by key and sort
dedup sorted
checkpoint Sort

ICFF = Indexed Compressed Flat Lookup File.
this file can do fast effective lookups. (Note: index is a hash.)

Components:
Write Block Compressed Lookups
Write Multiple Block Compressed lookups
- compresses data into blocks, indexes blocks, writes them out into file(s).

Lookup function: search index, finds block(s), read blocks from disk, uncompresses, searches inside the block for matching record(s).

ICFF take 10-times less disk space than normal uncompressed lookup files. Lookup time is fast.

Phases - are used to break the graph into pieces. Temporary files created during a phase will be deleted after its completion. Phases are used to effectively separately manage resource-consuming (memory, CPU, disk) parts of the application.

Checkpoints - created for recovery purposes. These are points where everything is written to disk. You can recover to the latest saved point - and rerun from it.

You can have phase breaks with or without checkpoints.

A new sandbox will have many directories: mp, dml, xfr, db, ... . xfr is a directory where you put files with extension .xfr containing your own custom functions (and then use : include "somepath/xfr/yourfile.xfr"). Usually XFR stores mapping.

1) Data Parallesim - data (partitionning of data into parallel streams for parallel processing).

2) Componnent Paralelism (execute simultaneously on different branches of the graph)

3) Pipeline (sequential).

Multi-File System

m_mkfs - create a multifile (m_mkfs ctrlfile mpfile1 ... mpfileN)
m_ls - list all the multifiles
m_rm - remove the multifile
m_cp - copy a multifile

m_mkdir - to add more directories to existing directory structure

• Each partition of a component uses: ~ 8 MB + max-core (if any)
• Add size of lookup files used in phase (if multiple components use same lookup only count it once)
• Multiply by degree of parallelism. Add up all components in a phase; that is how much memory is used in that phase.
• Select the largest-memory phase in the graph

SCAN
ROLLUP
SCANWITHROLLUP
Scan followed by Dedup sort and select the last

If we don't use any key in the sort component while using the dedup sort,
then the output depends on the keep parameter.

• first - only the first record
• last - only last record
• unique_only - there will be no records in the output file.
  
  

• use scan and filter
• m_dump <dml> <mfs file> -start 50 -end 75
• use next_in_sequence() function and filter by expression component (next_in_sequence() >50 && next_in_sequence() <75)

When you check out a project into your sandbox - you get project parameters. Once in your sandbox - you can refer to them as sandbox parameters.

You can not merge two ab initio graphs. You can use the ouput of one graph as input for another. You can also copy/paste the contents between graphs. See also about using .plan

• partitioning - dividing a single flow of records(serial file, mfs) into multiple flows.
• departitioning - removing partitionning (gather an merge component)
• re-partitioning - change the number of partitions (eg, from 2 to 4 flows)
  

• EME = Enterprise Metdata Environment. Functions (repository, version control, statistical analysis, dependency analysis). It is on the server side and holds all the projects (metadata of transformations, config info, source and target info: graph dml xfr ksh sql, etc..). This is where you checkin/checkout. /Project dir of EME contains common directories for all application sandboxes connected to it. It also helps in dependency analysis of codes. Ab Initio has series of air commands to manipulate repository objects.
• GDE = Graphical Devlopment Environment (on the client box)
• Co-operating sytem = Ab Initio server installed on top of native (unix) os on the server

fencing means job controlling on priority basis.
In AI it actually refers to customized phase breaking. A well fenced graph means no matter what is source data volume process will not cough in dead locks. It actually limits the number of simultaneous processes.

Fencing - changing a priority of a job
Phasing - managing the resources to avoid deadlocks.
For example, limiting the number of simultaneous processes
(by breaking the graph into phases, only 1 of which can run at any given time)

• AB_HOME - where co>operating system is installed
• AB_AIR_ROOT - default location for EME datastore
• sandboxes standard environment
• AI_SORT_MAX_CORE, AI_HOME, AI_SERIAL, AI_MFS, etc.
• from unix prompt: env | grep AI

Dynamic DML is used if the input metadata can change. Example: at different time different input files are recieved for processing which have different dml. in that case we can use flag in the dml and the flag is first read in the input file recieved and according to the flag its corresponding dml is used.

• fan out - partition component (increase parallelism)
• fan in departition component (decrease parallelism)

multi update executes SQL statements - it treats each input record as a completely separate piece of work.

• We can use Autosys, Control-M, or any other external scheduler.
• We can take care of dependencies in many ways. For example, if scripts should run sequentially, we can arrange for this in Autosys, or we can create a wrapper script and put there several sequential commands (nohup command1.ksh & ; nohup command2.ksh &; etc). We can even create a special graph in Ab Initio to execute individual scripts as needed.

Api and Utility modes in input table

• lookup file component. Functions: lookup, lookup_count, lookup_next, lookup_match, lookup_local.
• Lookups are always used with combination of the reformat components.

string_ltrim, string_lrtrim, string_substring, reinterpret_as, today(), now()

Ab Initio benefits: parallelism built in, mulitifile system, handles huge amounts of data, easy to build and run. Generates scripts which can be easily modified as needed )if something couldn't be done in ETL tool itself). The scripts can be easily scheduled using any external scheduler - and easily integrated with other systems.

Ab Initio doesn't require a dedicated administrator.

Ab Initio doesn't have built-in CDC capabilities (CDC = Change Data Capture).

Ab Initio allows to (attach error / reject files) to each transformation and capture and analyze the message and data separately (as opposed to Informatica which has just one huge log). Ab Initio provides immediate metrics for each component.

• Scenario 1 (preferred): we run query which joins 2 tables in DB and gives us the result in just 1 DB component.
• Scenario 2 (much slower): we use 2 database components, extract all data - and join them in Ab Initio.

Parameter showing how data is unevenly distributed between partitions.

skew = (partition size - avg.part.size)* 100 / (size of the largest partition)

.dbc - database configuration file (dbname, nodes, version user/pwd) - resides in the db directory

.cfg - any tyoe of config file. for example, remote connection config (name of remote server, user/pwd to connect to db, location of OS on remote machine, connection method). .cfg file resides in the config dir.

depth not equal data format error etc...

depth error : we get this error.. when two components connected together but does't match there layout

• Go parallel using partitionning. Roundrobin partitionning gives good balance.
• Use Multi-file system (MFS).
• Use Ad Hoc MFS to read many serial files in parallel, and use concat component.
• Once data is partitionned - do not switch it to serial and back. Repartition instead.
• Do not acceess large filess via NFS - use FTP instead
• use lookup local rather than lookup (especially for big lookups).
• Use rollup and Filter as soon as possible to reduce number of records. Ideally do it in the source (database ?) before you get the data.
• Remove unnecessary components. For example, instead of using filter by exp, you can implement the same function in reformat/Join/Rollup. Another example - when joining data from 2 files, use union function instead of adding an additional component for removing duplicates.
• use gather instead of concatenate.
• it is faster to do a sort after a partitino, than to do a sort before a partition.
• try to avoid using a join with the "db" component.
• when getting data from database - make sure your queries are fast (use indexes, etc.). If possible, do necessary selection / aggregation / sorting in the database before getting data into Ab Initio.
• tune Max_core for Optimal performance (for sort depends on the size of the input file).
• Note - If in-memory join cannot fit its non-driving inputs in the provided MAX-CORE, then it will drop all the inputs to disk and in-memory does not make sence.
• Using phase breaks let you allocate more memory in individual components - thus improving performance.
• Use checkpoint after sort to land data on disk
• Use Join and rollup in-memory feature
• When joining very small dataset to a very large dataset it is more efficient to broadcast the small dataset to MFS using broadcast component, or use the small file as lookup. But for large dataset don't use broadcast as a partitioner.
• Use Ab Initio layout instead of database default to achieve parallel loads
• Change AB_REPORT parameter to increased monitoring duration
• Use catalogs for reusability
• Components like join/ rollup should have the option "Input must be sorted"
  if they are placed after a sort component.
• minimize number of sort components. Minimize usage of sorted join component, and if possible replace them by in-memory join/hash join. Use only required fields in the sort reformat join components. Use "Sort within Groups" instead of just Sort when data was already presorted.
• Use phasing/flow buffers in case of merge sorted joins
• Minimize the use of regular expression functions like re_index in the transfer functions
• Avoid repartitioning of data unnecessarily. When splitting records into more than two flows, use Reformat rather than Broadcast component.
• For joining records from 2 flows use Concatenate component ONLY when there is a need to follow some specific order in joining records. If no order is required then it is preferable to use Gather component.
• Instead of putting many Reformat components consecutively, use output indexes parameter in the first Reformat component and mention the condition there.
  
  

• Delta table maintain the sequencer of each data table.
• Master (or base) table - a table on tp of which we create a view
  

• Reformat - deriving new data by adding/dropping fields
• Redefine format - rename fields

• The prerequisit for using sortwithingroup is that the data is already sorted by the major key. sortwithingroup outputs the data once it has finished reading the major key group. It is like an implicit phase.

Define a Formal Keyword Parameter of type string. For example, you call it FilterCondition, and you want it to do filtering on COUNT > 0 . Also in your graph in your "Filter by expression" Component enter following condition: $FilterCondition

Now on your command line or in wrapper script give the following command
     YourGraphname.ksh -FilterCondition COUNT > 0

#!/bin/ksh

#Running the set up script on   enviornment
typeset PROJ_DIR $(cd $(dirname $0)/..; pwd) 
.   $PROJ_DIR/ab_project_setup.ksh $PROJ_DIR
#Exporting the script parameter1 to   INPUT_FILE_NAME
if [ $# -ne 2 ];
then
  INPUT_FILE_PARAMETER_1   $1
  INPUT_FILE_PARAMETER_2 $2
  # This grpah is using the input   file
  cd $AI_RUN
  ./my_graph1.ksh $INPUT_FILE_PARAMETER_1
  # This graph also is using the input   file.
  ./my_graph2.ksh $INPUT_FILE_PARAMETER_2
  exit 0;
else
  echo Insufficient   parameters 
  exit 1;
fi
-------------------------------------
#!/bin/ksh

#Running the set up script on enviornment
typeset PROJ_DIR $(cd $(dirname $0)/..; pwd) 
. $PROJ_DIR/ab_project_setup.ksh $PROJ_DIR

#Exporting the script parameter1 to INPUT_FILE_NAME
export INPUT_FILE_NAME $1

# This grpah is using the input file
cd $AI_RUN
./my_graph1.ksh

# This graph also is using the input file.
./my_graph2.ksh

exit 0;

use conditional dml where you can separate detail from header and trailer. For validations use reformat with count :3 (out0:header out1:detail out2:trailer.)

• first method: in GDE go to RUN > Execute Command - and run m_mkfs c:control c:dp1 c:dp2 c:dp3 c:dp4
• second method: double-click on the file component, and in ports tab double-click on partitions - there you can enter the number of partitions.

There are many ways to create a surrogate key. For example, you can use next_in_sequence() function in your transform. Or you can use "Assign key values" component. Or you can write a stored procedure - and call it.

Note: if you use partitions, then do something like this:

(next_in_sequence()-1)*no_of_partition()+this_partition()

This is a config file for ab initio - in user's home directory and in $AB_HOME/Config. It sets abinitio home path, configuration variables (AB_WORK_DIR, AB_DATA_DIR, etc.), login info (id, encrypted password), login methods for hosts for execution (like EME host, etc.), etc.

your ksh init file ( environment, aliases, path variables, history file settings, command prompt settings, etc.)

• input file / output file
• input table / output table
• lookup / lookup_local
• reformat
• gather / concatenate
• join
• runsql
• join with db
• compression components
• filter by expression
• sort (single or multiple keys)
• rollup
• trash
• partition by expression / partition by key

This is basically an Oracle question - regarding SQLLDR (SQL Loader) utility. Conventional load - using insert statements. All triggers will fire, all contraints will be checked, all indexes will be updated.

Direct load - data is written directly block by block. Can load into specific partition. Some constraints are checked, indexes may be disabled - need to specify native options to skip index maintenance.

abinitio online help gives 3 examples of joins: inner join, outer join, and semi join.

• for inner join 'record_requiredN' parameter is true for all "in" ports.
• for outer join it is false for all the "in" ports.
• for semi join it is true for both port (like InnerJoin), but the dedup option is set only on one side