QuantumLeap is a REST service for storing, querying and retrieving NGSI v2 and NGSI-LD (experimental support) spatial-temporal data. QuantumLeap converts NGSI semi-structured data into tabular format and stores it in a time-series database, associating each database record with a time index and, if present in the NGSI data, a location on Earth. REST clients can then retrieve NGSI entities by filtering entity sets through time ranges and spatial operators. Note that, from the client's stand point, these queries are defined on NGSI entities as opposed to database tables. However, the query functionality available through the REST interface is quite basic and most complex queries typically require clients to use the database directly.
The REST API specification, dubbed NGSI-TSDB, which QuantumLeap implements has been defined with the goal of providing a database-agnostic REST interface for the storage, querying and retrieval of NGSI entity time series that could be as close as possible to the NGSI specification itself. Thus NGSI-TSDB provides a uniform and familiar (to FIWARE developers) mechanism to access time series data which allows implementing services such as QuantumLeap to transparently support multiple database back ends. In fact, presently QuantumLeap supports both CrateDB and Timescale as back end databases.
temporal attributes are not currently supported (#395); what is relevant here is that this attributes are used to create the time index of the series
other attributes may be added as well in future (not a priority probably, so may not be tackled any time #396)
context is currently not stored.
query endpoints returns NGSIv2 data types.
NGSI-LD temporal queries seem to have a semantic that implies that only numeric values are tracked in time series. This was never the case for QL that trace over time any attribute (also not numeric ones), since they may change as well.
NGSI-LD semantics also seem to track values over time of single attributes. QL to enable to retrieve full entity values in a given point in time stores the whole entity in a single table (this avoids the need for JOINs that are notoriously time consuming - but on the other hand generates more sparse data). In doing so, we create for the entity a single time index, this is due to the fact that while different dateTime attributes can be defined and hence queried, only one can be used to index time series in all modern timeseries DB (to achieve performance). This imply that we have a policy to compute such time index (either custom and referring to an attribute of the entity, or using the "latest" time metadata linked to the entity or to an attribute). The issue is that if the notification payload sent to QL includes all attributes, also not update ones, QL will "timestamp" all values (also old ones) with that timestamp.
This means that the ability to track a specific value of an attribute in a point in time depends on the actual notification.
In short, given that we aim to ensure both forward compatibility (data store as NGSIv2 can be queried in future as NGSI-LD) and backward compatibility (data store as NGSI-LD can be queried as NGSIv2), implementing NGSI-LD temporal api, may not be 100% compliant with the specs.
Relation to STH Comet
Although QuantumLeap and FIWARE STH Comet share similar goals, Comet doesn't support multiple database back ends (only MongoDB is available) and doesn't support NGSI v2 either. While Comet per se is a fine piece of software, some of the needs and assumptions that prompted its developments are no longer current. QuantumLeap started out as an exploration of an alternative way to make historical data available to the FIWARE ecosystem without committing to a specific database back end.
Typically QuantumLeap acquires IoT data, in the form of NGSI entities, from a FIWARE IoT Agent layer indirectly through NGSI notifications set up upfront with the context broker, Orion. (We assume the reader is familiar with the NGSI publish-subscribe mechanism described in the Notification Messages and Subscriptions sections of the NGSI specification.) As mentioned earlier, incoming NGSI entities are converted to database records and stored in one of the configured time series database back ends---typically, a database cluster. Often visualisation tools such as Grafana are deployed too in order to visualise the time series data that QuantumLeap stores in the database. The below diagram illustrates relationships and interactions among these systems in a typical QuantumLeap deployment scenario.
In order for QuantumLeap to receive data from Orion, a client creates a subscription in Orion specifying which entities should be notified when a change happens (1). (You can read more about setting up subscriptions in the Orion Subscription section of the QuantumLeap manual.)
From this point on, when Agents in the IoT layer push data to the context broker (2), if the data pertains to entities pinpointed by the client subscription, Orion forwards the data to QuantumLeap by POSTing NGSI entities to QuantumLeap's notify end point (3).
QuantumLeap's Reporter component parses and validates POSTed data. Additionally, if geo-coding is configured, the Reporter invokes the Geocoder component to harmonise the location representation of the notified entities, which involves looking up geographic information in OpenStreetMap (OSM). At this stage, depending on the deployed mode you selected, data are immediately processed or stored in the cache for later processing. In the first case, the Reporter passes on the validated and harmonised NGSI entities to a Translator component. In the second case, the Reporter stores on the validated and harmonised NGSI entities to the Cache component, that is acting as a message queue. The Worker component will read pending messages to be processed and will pass them to a Translator component. The Admin Guide contains more details about the work queue.
Translator convert NGSI entities to tabular format and persist them as time series records in the database. There is a Translator component in correspondence of each supported database back end - see section below. Depending on the configuration, a specific Translator is used.
When a client queries the REST API to retrieve NGSI entities (4), the Reporter and Translator interact to turn the Web query into a SQL query with spatial and temporal clauses, retrieve the database records and convert them back to the NSGI entity time series eventually returned to the client. As noted earlier, the query functionality available through the REST interface is quite basic: QuantumLeap supports filtering by time range, geographical queries as defined by the NGSI specification and simple aggregate functions such as averages. Other than that, QuantumLeap also supports deleting historical records but note that presently it does not implement in full the NGSI-TSDB specification - please refer to the REST API specification for the details.
Finally, the diagram shows a Dashboard querying the database directly in order to visualise time series for a Web client (5). In principle, it should be possible to develop a Dashboard can also query the QuantumLeap REST API instead of the database which would shield visualisation tools from QuantumLeap internals.
Database Back Ends
One guiding principle in QuantumLeap design has been the ability to use multiple time series databases. This design choice is justified by the fact that a database product may be more suitable than another depending on circumstances at hand. Currently QuantumLeap can be used with both CrateDB and Timescale.
The Database Selection section of this manual explains how to configure QuantumLeap to use one of the available database back ends.
CrateDB back end
CrateDB is the default back end. It is easy to scale thanks to its shared-nothing architecture which lends itself well to containerisation so it is relatively easy to manage a containerised CrateDB database cluster, e.g. using Kubernetes. Moreover, CrateDB uses SQL to query data, with built-in extensions for temporal and geographical queries. CrateDB offers as well a Postgres API, making simpler its integration. For example, you can leverage Grafana PostgreSQL plugin to visualise time series stored in CrateDB.
QuantumLeap stores NGSI entities in CrateDB using the
------------------------- --------------- | CrateDB | <----- | QuantumLeap |-----O notify ------------------------- ---------------
Timescale back end
Timescale is another time series databases that can be used with QuantumLeap as a back end to store NGSI entity time series. Indeed, QuantumLeap provides full support for storing NGSI entities in Timescale, including geographical features (encoded as GeoJSON or NGSI Simple Location Format), structured types and arrays.
------------------------- | Timescale PostGIS | --------------- | --------------------- | <----- | QuantumLeap |-----O notify | Postgres | --------------- -------------------------
PostgreSQL is a rock-solid, battle-tested, open source database, and its PostGIS extension provides excellent support for advanced spatial functionality while the Timescale extension has fairly robust support for time series data. The mechanics of converting an NGSI entity to tabular format stay pretty much the same as in the Crate back end except for a few improvements:
NGSI arrays are stored as (indexable & queryable) JSON as opposed to the flat array of strings in the Crate back end.
GeoJSON and NGSI Simple Location Format attributes are stored as spatial data that can be indexed and queried - full support for spatial attributes is still patchy in the Crate back end.
test_timescale_insert.py file in the QuantumLeap source base
contains quite a number of examples of how NGSI data are stored in
Note: querying & retrieving database
At the moment, QuantumLeap implement experimental querying of data through the QuantumLeap REST API. This means that while REST API on top of CrateDB have been tested in production, this is not the case for Timescale.
Cache Back End
To reduce queries to databases or to geocoding APIs, QuantumLeap leverages a cache. The only cache backend supported so far is Redis. Caching support for queries to databases is experimental.
-------------------- --------------- | DB | ------ | QuantumLeap |-----O notify -------------------- --------------- | | --------------- | Redis | ---------------
As of today, the query caching stores:
Version of CrateDB. Different version of CrateDB supports different SQL dialects, so at each request we check which version of CrateDB we are using. By caching this information, each thread will ask this information only once. Of course this could be passed as variable, but then live updates would require QL down time. Currently, you can update from a Crate version to another with almost zero down time (except the one caused by Crate not being in ready state), you would need only to clear the key
cratefrom redis cache. TTL in this case is 1 hour, i.e. after one hour version will be checked again against CrateDB.
Metadata table. The metadata table is used to store information about the mapping between original NGSI attributes (including type) to db column names. Basically this information is required to perform "consistent" data injection and to correctly NGSI type retrieved attributes by queries. Given concurrency due to the support of multithread and ha deployment, cache in this case has by default a shorter TTL (60 sec). Cache is anyhow re-set every time a change to Metadata table occurs (e.g. in case the incoming payload include a new entityType or a new attribute for an existing entityType). Metadata for a specific entityType are removed only if a entityType is dropped, not in case all its values are removed.