Skip to end of metadata
Go to start of metadata

You are viewing an old version of this page. View the current version.

Compare with Current View Page History

« Previous Version 7 Next »

Executive Summary


The Squirro Pipeline 2.0 allows you to import large amounts of data in such a resource-friendly way that Squirro remains responsive even during high volume data inflow. This document describes the steps you can follow to use Pipeline 2.0 to import large amounts of data into Squirro installations versioned 2.5.1 or 2.5.2. The Squirro Pipeline 2.0 is available starting with Squirro version 2.5.1, and will become the default way to import data into Squirro with release 2.5.3 later in 2017.

When to use Pipeline 2.0

Whenever you have hundreds of millions of documents to import or if you don't need all the built-in enrichments that come at a processing cost and slow down data ingestion, Pipeline 2.0 is for you. A key feature of Pipeline 2.0 is that you would "opt in" to enrichments rather than paying the price of running all enrichments.

Because the Pipeline 2.0 can run alongside Pipeline 1.0, you have a couple of options:

  1. Enable Pipeline 2.0 Squirro Server-wide: This is the best option if you are starting brand-new Squirro projects on a dedicated Squirro installation and need to ingest large data volumes.
  2. Enable Pipeline 2.0 for individual projects or subscriptions: This may be an option for existing projects to which you may be planning to augment with additional, large data subscriptions.

Because Pipeline 2.0 does not run any of the built-in enrichments by default (until you opt in each enrichment individually), switching existing subscriptions may have unintended side-effects. This is because your application and users have likely come to rely on the behavior introduced by the enrichments. In such cases we recommend contacting Squirro support if you are interested in speeding up the import of existing data subscriptions.

Enabling Pipeline 2.0 Squirro Server wide

Pipeline 2.0 relies on a file system to queue data in batches before Squirro inserts the data into ElasticSearch in bulk. Options for storing this temporary data files are:

  1. Local file system: This is the default out-of-the-box configuration with queued data placed under /var/lib/squirro/inputstream/ . We recommend using Raid-1 or some other form of redundancy to guard against loss of this temporary data. The IO access generated by the Pipeline 2.0 consists of sequential reads and writes consistent with log structured storage in which data is added sequentially "to the end" of the queue and read and deleted "from the beginning" of the queue as data is inserted into ElasticSearch.

    Note in the case of multiple Squirro Cluster Node installations that the queue and file system are implicitly "sharded" with a different subset of the data going to different servers and disks. This helps scale the data capacity of the Pipeline 2.0 much like you can scale ElasticSearch by adding more Indexing Servers.

  2. Amazon Elastic Block Storage: For cloud installations hosted in AWS, Amazon EBS is a suitable choice.
  3. Network attached storage: NAS is expensive, so although network attached storage will work, it is likely not the first choice.

Internally Pipeline 2.0 is known as "Ingester", or specifically the sqingesterd service. To have all new subscriptions import data via the Pipeline 2.0, configure the Squirro API Provider service (the service that all Squirro data imports interact with), like so:

/etc/squirro/provider.ini
[provider]
# new pipeline configs:
# processing_mode controls the transition to the ingester ("new pipeline")
# modes: legacy (all sources go through only old pipeline and bulk pipeline),
#        tee (all sources go through legacy & new pipelines for "shadow-testing"),
#        default_legacy (by default sources go through legacy
#            pipelines, but can be overridden in source config to new pipeline),
#        default_ingester (by default sources go through new pipeline, but can be
#            overridden in source config to legacy pipelines),
#        ingester (all sources exclusively go through new pipeline)
#
processing_mode = default_ingester

Ensure that /var/lib/squirro/inputstream has enough space. At the minimum we require 10 GB is space. The following setting in /etc/squirro/common.ini ensures that the Pipeline 2.0 does not fill up the disk space and leaves at least 10 GB of space for other users of the same disk volume.

[content_filesystem_stream]
# directories for pipeline 2.0 queued data
source_metadata_directory = /var/lib/squirro/inputstream
data_directories = %(source_metadata_directory)s

# number of gigabytes to require at least to continue writing to the file system
back_off_when_data_disk_space_falls_below_in_gigabytes = 10
back_off_when_metadata_disk_space_falls_below_in_gigabytes = %(back_off_when_data_disk_space_falls_below_in_gigabytes)s

If you elect not to use the local file system, change source_metadata_directory and data_directories accordingly.

Enabling Pipeline 2.0 for Individual Subscriptions

The same as in the previous section, ensure that there is enough space and sequential IO capacity for temporarily queued data files.

Then follow the step described under Processing Config to choose the Pipeline 2.0 when creating a subscription.

We are in the process of enabling our Data Import Tools to allow you to more conveniently choose the Pipeline 2.0 starting shortly with the Squirro Data Loader.


  • No labels