Maintaining Spatial Data Infrastructures (SDIs) using distributed task queues
Paolo Corti and Ben LewisHarvard Center for Geographic Analysis
2017 FOSS4GBoston
Background
Harvard Center for Geographic Analysis
• WorldMap http://worldmap.harvard.edu– Biggest GeoNode instance on the planet– https://github.com/cga-harvard/cga-worldmap
• HHypermap http://hh.worldmap.harvard.edu – Map service registry– https://github.com/cga-harvard/HHypermap
NoteBillion Object Platform (BOP)
https://github.com/cga-harvard/hhypermap-bop
Demo of WorldMap / HHypermap
The need for an asynchronous processor
In WorldMap and HHypermap there are operations run by users which are time consuming and cannot be handled in the context of a web request
● Harvest the metadata of a service and its layers● Synchronize the metadata of a new or updated layer to the search
engine● Feed a gazetteer when a new layer is uploaded or updated● Upload a spatial datasets to the server● Create a new layer using a table join
HTTP request/response cycle must be fast● In web applications the HTTP
request/response cycle can be synchronous as long as there are very quick interactions between the client and the server
● unfortunately there are cases when the cycle become slower
● In these situations the best practice for a web application is to process asynchronously these tasks using a task queue
Task Queues
Asynchronous processing in a web application can be delegated to a task queue, which is a system for parallel execution of tasks in a non-blocking fashion
Asynchronous processing model
Asynchronous processing model
● The asynchronous processing model is composed by services that produce processing tasks (producers) and by services which consume and process these tasks (consumers) accordingly
● A message queue is a broker which facilitates message passing by providing a protocol or interface which other services can access. Work can be distributed across threads or machines
● In the context of a web application the producer is the client application that creates messages based on the user interaction. The consumer is a daemon process that can consume the messages and run the needed process
Glossary● Task Queue: a system for parallel execution of tasks in a non-blocking
fashion● Broker or Message Queue: provides a protocol or interface for messages
exchanging between different services and applications● Producer: the code that places the tasks to be executed later in the broker● Consumer or Worker: takes tasks from the broker and process them● Exchange: takes a message from a producer and route it to zero or more
queues (messages routing)
Tasks must be consumed faster than being produced. If not, add more workers
Use cases for task queues
● in web applications some process is taking too much time and must be processed asynchronously
● heterogeneous applications/services in a given system architecture need an easy way to reliably communicate between each other
● periodic operations (vs crontab)● a way of parallelizing tasks in multi processors● monitor processes and analyze failing tasks (and execute
them again)
Typical use cases for a task queue in a web application
● Thumbnails generation● Sending bulk email● Fetching large amounts of data from APIs● Performing time-intensive calculations● Expensive queries● Search engine index synchronization● Interaction with another application/service● Replacing cron jobs (backups, maintenance, etc…)
Typical use cases for a task queue in a GIS Portal/SDI
● Upload a shapefile to the server (GeoNode)● Thumbnails generation for layers and maps (GeoNode)● OGC services harvesting (Harvard Hypermap)● Geoprocessing operations● Geospatial data maintenance
Producer, broker and consumer architecture
Producer
Consumer
ProducerBroker
Consumer
Producer
Broker
ConsumerProducer
BrokerConsumer
Producer
Message brokers implementations
Most of them are open source!
● RabbitMQ (AMQP, STOMP, JMS)● Apache ActiveMQ (STOMP, JMS)● Amazon Simple Queue Service (JMS)● Apache Kafka
Several standard protocols:
● AMQP, STOMP, JMS, MSMQ (Microsoft .NET)
Tasks (Jobs) queues implementations
● Celery (RabbitMQ, Redis, Amazon SQS, Zookeeper)● Redis Queue (Redis)● Resque (Redis)● Kue (Redis)
And many others!
Celery● asynchronous task queue based on distributed message
passing● focused on real-time operation, but supports scheduling
as well● the execution units, called tasks, are executed
concurrently on a single or more worker servers● it supports many message brokers (RabbitMQ, Redis,
MongoDB, CouchDB, ...)● written in Python but it can operate with other languages● great integration with Django!● great monitoring tools (Flower, django-celery-results)
RabbitMQ
● RabbitMQ is a message broker: it accepts and forwards messages
● most widely deployed open source broker (35k+ deployments)
● support many message protocols● supported by many operating systems and
languages● Written in Erlang
Architecture of Celery/RabbitMQ
https://tests4geeks.com/python-celery-rabbitmq-tutorial/
A real use case: Harvard HypermapHHypermap (Harvard Hypermap) Registry is a platform that manages OWS, Esri REST, and other types of map service harvesting, and orchestration and maintains uptime statistics for services and layers. Where possible, layers are cached by MapProxy.
HHypermap provides thousands of remote layers to WorldMap users
Harvard HypermapWorldMap Architecture
HHypermap interface
Need for a task queue
SLOW!!!
Producer
Is the code that places the tasks to be executed later in the broker
Celery messages
Consumer
Takes tasks from the broker and process them in a worker
Replacing cron jobs
Replacing cron jobs
Workers and threads with htop
Monitoring
Monitoring a task
Thanks!
Question and Answer