Good practices for ETL

Four-layered approach for ETL architecture design

Functional layer: Core functional ETL processing (extract, transform, and load).

Operational management layer: Job-stream definition and management, parameters, scheduling, monitoring, communication and alerting.

Audit, balance and control (ABC) layer: Job-execution statistics, balancing and controls, rejects- and error-handling, codes management.

Utility layer: Common components supporting all other layers.

Use file-based ETL processing where possible

Storage costs relatively little

Intermediate files serve multiple purposes

Used for testing and debugging

Used for restart and recover processing

Used to calculate control statistics

Helps to reduce dependencies - enables modular programming.

Allows flexibility for job-execution and -scheduling

Better performance if coded properly, and can take advantage of parallel processing capabilities when the need arises.

Use data-driven methods and minimize custom ETL coding

Parameter-driven jobs, functions, and job-control

Code definitions and mapping in database

Consideration for data-driven tables to support more complex code-mappings and business-rule 

application.

Qualities of a good ETL architecture design

Performance

Scalable

Migratable

Recoverable (run_id, ...)

Operable (completion-codes for phases, re-running from checkpoints, etc.)

Auditable (in two dimensions: business requirements and technical troubleshooting)

Rerunnability, recoverability in ETL

Data warehousing procedures usually subdivide a big ETL process into smaller pieces running sequentially or in parallel. To keep track of data flows, it makes sense to tag each data row with "row_id", and tag each piece of the process with "run_id". In case of a failure, having these IDs will help to roll back and rerun the failed piece.

Best practice also calls for "checkpoints", which are states when certain phases of the process are completed. Once at a checkpoint, it is a good idea to write everything to disk, clean out some temporary files, log the state, and so on.

Parallel processing in ETL

A recent development in ETL software is the implementation of parallel processing. This has enabled a number of methods to improve overall performance of ETL processes when dealing with large volumes of data.

ETL applications implement three main types of parallelism:

Data: By splitting a single sequential file into smaller data files to provide parallel access.

Pipeline: Allowing the simultaneous running of several components on the same data stream. For example: looking up a value on record 1 at the same time as adding two fields on record 2.

Component: The simultaneous running of multiple processes on different data streams in the same job, for example, sorting one input file while removing duplicates on another file.

All three types of parallelism usually operate combined in a single job.

An additional difficulty comes with making sure that the data being uploaded is relatively consistent. Because multiple source databases may have different update cycles (some may be updated every few minutes, while others may take days or weeks), an ETL system may be required to hold back certain data until all sources are synchronized. Likewise, where a warehouse may have to be reconciled to the contents in a source system or with the general ledger, establishing synchronization and reconciliation points becomes necessary.

Performance of ETL

ETL vendors benchmark their record-systems at multiple TB (terabytes) per hour (or ~1 GB per second) using powerful servers with multiple CPUs, multiple hard drives, multiple gigabit-network connections, and lots of memory.

In real life, the slowest part of an ETL process usually occurs in the database load phase. Databases may perform slowly because they have to take care of concurrency, integrity maintenance, and indexes. Thus, for better performance, it may make sense to do most of the ETL processing outside of the database, and to use bulk load operations whenever possible. Still, even using bulk operations, database access is usually the bottleneck in the ETL process. Here are some common methods used to increase performance:

Partition tables (and indices). Try to keep partitions similar in size (watch for null values which can skew the partitioning).

Do all validation in the ETL layer before the load. Disable integrity checking (disable constraint ...) in the target database tables during the load.

Disable triggers (disable trigger ...) in the target database tables during the load. Simulate their effect as a separate step.

Generate IDs in the ETL layer (not in the database).

Drop the indexes (on a table or partition) before the load - and recreate them after the load (SQL: drop index ...; create index ...).

Use parallel bulk load when possible — works well when the table is partitioned or there are no indexes. Note: attempt to do parallel loads into the same table (partition) usually causes locks — if not on the data rows, then on indexes.

If a requirement exists to do insertions, updates, or deletions, find out which rows should be processed in which way in the ETL layer, and then process these three operations in the database separately. You often can do bulk load for inserts, but updates and deletes commonly go through an API (using SQL).

Whether to do certain operations in the database or outside may involve a trade-off. For example, removing duplicates using distinct may be slow in the database; thus, it makes sense to do it outside. On the other side, if using distinct will significantly (x100) decrease the number of rows to be extracted, then it makes sense to remove duplications as early as possible in the database before unloading data.

A common source of problems in ETL is a big number of dependencies among ETL jobs. For example, job "B" cannot start while job "A" is not finished. You can usually achieve better performance by visualizing all processes on a graph, and trying to reduce the graph making maximum use of parallelism, and making "chains" of consecutive processing as short as possible. 

Again, partitioning of big tables and of their indexes can really help.

Another common issue occurs when the data is spread between several databases, and processing is done in those databases sequentially. Sometimes database replication may be involved as a method of copying data between databases - and this can significantly slow down the whole process. The common solution is to reduce the processing graph to only three layers:

Sources

Central ETL layer

Targets

This allows processing to take maximum advantage of parallel processing. For example, if you need to load data into two databases, you can run the loads in parallel (instead of loading into 1st - and then replicating into the 2nd).

Of course, sometimes processing must take place sequentially. For example, you usually need to get dimensional (reference) data before you can get and validate the rows for main "fact" tables.

Challenges in ETL

ETL processes can involve considerable complexity, and significant operational problems can occur with improperly designed ETL systems.

The range of data values or data quality in an operational system may exceed the expectations of designers at the time validation and transformation rules are specified. Data profiling of a source during data analysis is recommended to identify the data conditions that will need to be managed by transform rules specifications. This will lead to an amendment of validation rules explicitly and implicitly implemented in the ETL process.

Data warehouses typically grow asynchronously, fed by a variety of sources which all serve a different purpose, resulting in, for example, different reference data. ETL is a key process to bring heterogeneous and asynchronous source extracts to a homogeneous environment.

Design analysts should establish the scalability of an ETL system across the lifetime of its usage. This includes understanding the volumes of data that will have to be processed within service level agreements. The time available to extract from source systems may change, which may mean the same amount of data may have to be processed in less time. Some ETL systems have to scale to process terabytes of data to update data warehouses with tens of terabytes of data. Increasing volumes of data may require designs that can scale from daily batch to multiple-day microbatch to integration with message queues or real-time change-data capture for continuous transformation and update.

Real-life ETL cycle

The typical real-life ETL cycle consists of the following execution steps:

1. Cycle initiation

2.Build reference data

3.Extract (from sources)

4.Validate

5.Transform (clean, apply business rules, check for data integrity, create aggregates)

6.Stage (load into staging tables, if used)

7.Audit reports (for example, on compliance with business rules. Also, in case of failure, helps to diagnose/repair)

8.Publish (to target tables)

9.Archive

10.Clean up

Load Concept

The load phase loads the data into the end target, usually the data warehouse (DW). Depending on the requirements of the organization, this process varies widely. Some data warehouses may overwrite existing information with cumulative, updated data every week, while other DW (or even other parts of the same DW) may add new data in a historized form, for example, hourly. The timing and scope to replace or append are strategic design choices dependent on the time available and the business needs. More complex systems can maintain a history and audit trail of all changes to the data loaded in the DW.

As the load phase interacts with a database, the constraints defined in the database schema — as well as in triggers activated upon data load — apply (for example, uniqueness, referential integrity, mandatory fields), which also contribute to the overall data quality performance of the ETL process.

Transform Concept

The transform stage applies to a series of rules or functions to the extracted data from the source to derive the data for loading into the end target. Some data sources will require very little or even no manipulation of data. In other cases, one or more of the following transformations types to meet the business and technical needs of the end target may be required:

Selecting only certain columns to load (or selecting null columns not to load)

Translating coded values (e.g., if the source system stores 1 for male and 2 for female, but the warehouse stores M for male and F for female), this calls for automated data cleansing; no manual cleansing occurs during ETL

Encoding free-form values (e.g., mapping "Male" to "1" and "Mr" to M)

Deriving a new calculated value (e.g., sale_amount = qty * unit_price)

Filtering

Sorting

Joining data from multiple sources (e.g., lookup, merge)

Aggregation (for example, rollup - summarizing multiple rows of data - total sales for each store, and for each region, etc.)

Generating surrogate-key values

Transposing or pivoting (turning multiple columns into multiple rows or vice versa)

Splitting a column into multiple columns (e.g., putting a comma-separated list specified as a string in one column as individual values in different columns)

Applying any form of simple or complex data validation. If validation fails, it may result in a full, partial or no rejection of the data, and thus none, some or all the data is handed over to the next step, depending on the rule design and exception handling. Many of the above transformations may result in exceptions, for example, when a code translation parses an unknown code in the extracted data.

Extract Concepts

The first part of an ETL process involves extracting the data from the source systems. Most data warehousing projects consolidate data from different source systems. Each separate system may also use a different data organization / format. Common data source formats are relational databases and flat files, but may include non-relational database structures such as Information Management System (IMS) or other data structures such as Virtual Storage Access Method (VSAM) or Indexed Sequential Access Method (ISAM), or even fetching from outside sources such as web spidering or screen-scraping. Extraction converts the data into a format for transformation processing.

An intrinsic part of the extraction involves the parsing of extracted data, resulting in a check if the data meets an expected pattern or structure. If not, the data may be rejected entirely.

ETL (Extract, Transform, Load) Basics

Extract, transform, and load (ETL) in database usage and especially in data warehousing involves:

Extracting data from outside sources

Transforming it to fit operational needs (which can include quality levels)

Loading it into the end target (database or data warehouse)

The advantages of efficient and consistent databases make ETL very important as the way data actually gets loaded.

This article discusses ETL in the context of a data warehouse, whereas the term ETL can in fact refer to a process that loads any database.

ETL can also function to integrate contemporary data with legacy systems.

Usually ETL implementations store an audit trail on positive and negative process runs. In almost all designs, this audit trail does not give the level of granularity which would allow a DBA to reproduce the ETL's result in the absence of the raw data.