Release notes v2.7#

This release introduced several improvements and new features. A few bugs were fixed, some outputs were changed and there were improvements to the installers, to the Web User Interface (WebUI) and to the engine itself.

Nearly 100 pull requests were closed. For the complete list of changes, please see the changelog:


We discovered that the disaggregation calculation was not working on a cluster unless a shared_dir was specified in openquake.cfg. This has been solved. Also, the rupture filtering was not applied in the disaggregation phase and this has been fixed too.

The procedure to export the hazard curves in .hdf5 format with the command oq export hcurves/all had some bugs and has been deprecated. Instead, we recommend the command oq extract hazard/all which is more reliable and exports also the hazard maps and uniform hazard spectra.

The command oq to_shapefile has been extended to work with source models in format NRML 0.5. The only missing feature is the support for multiPointSources.

On the hazardlib side, four new GMPEs entered:

  • Bindi et al. (2017)

  • Zhao et al. (2016)

  • Derras et al. (2014)

  • Pankow and Pechmann (2004)


Lots of changes entered on the risk side, mostly with the intent of supporting the OpenQuake QGIS plugin. The idea is to remove some logic (and some outputs) from the engine and to have the QGIS plugin produce those outputs instead. This applies in particular to the aggregated outputs for the risk calculators. We have deprecated them from the engine and they will be removed in future releases.

We added an aggregation API which is used by the QGIS plugin to perform the aggregation on-the-fly. The API can also be used by third party applications. For people not using the QGIS plugin there is also a command-line interface providing the same aggregation features. No features are lost and actually we have more features than before, including rather sophisticated aggregations by tag. Such features are included in this release, but they are experimental for the moment and they will be documented in the next release.

We added a node <tagNames>...</tagNames> in the exposure which is mandatory if you want to use tags for the assets (a feature introduced in the engine 2.6). The Input Preparation Toolkit will soon support the generation of exposures with asset tags.

We also changed the event loss table CSV exporter: now all realizations are exported in a single file containing an additional column which is the realization index. For the rest, the output is unchanged. This CSV output for the aggregate event loss table has been deprecated, anyway, because in the future it will likely disappear and be replaced with an API which will support loss aggregation based on the asset tags and taxonomies.

The scenario_damage calculator has been optimized significantly. In particular, the ScenarioDamage demo is now 17 times faster than before in the risk part of the calculation. The trick was to vectorize the calculation of the damage fractions, possible now that the fragility functions accept arrays as inputs (an user-requested feature, thanks to Hyeuk Ryu for the suggestion).

Another long-standing bug related to the classical_damage calculator has now been fixed: if there are some probabilities of exceedence (PoEs) exactly equal to 1.0 in the hazard curves, now the classical_damage calculator uses a cutoff and does not fail with a log(1 - PoE) going to infinity.

If a user provides an invalid vulnerability function with a probability mass function (PMF), now a much better error message is raised, including the line number where the mistake has been made.

There were several bugs in the new and experimental gmf_ebrisk calculator: all the reported bugs have been fixed now. The calculator still works only for ground motion fields of kind scenario, but there is a plan to extend it so that it can use the GMFs generated by an event based calculation as input. This calculator will be documented in the manual for the next release of the engine.

There was a small bug in the classical_risk calculator: while the loss curves where computed and exportable with oq export, they were not listed in the outputs of the engine. Now they are and you can export them with the oq engine export command. oq export is still needed if you want to export the individual realizations, since the engine only exposes the statistics.

There has been a lot of refactoring going on with the risk calculators: the final goal is to make the risk calculators easy to use programmatically. We are still at the beginning of the road, but significant steps in this direction have been taken.


The integration of the WebUI with the standalone tools has been improved and now they are automatically visible in the WebUI if installed, which is the case for the virtual machines we distribute and for the Windows installer, but not for the Linux packages.

The WebUI now support groups of user, i.e. you can restrict the access as you want. The admin user sees a link in the WebUI from which she can directly configure the groups by using the Django admin interface. System administrators interested in using this feature should read By default, no authentication is enabled and everything is visible to everybody.

The WebUI displays a message if you are running an obsolete version of the engine. The same message appears also in the command-line every time you run a calculation with an obsolete version of the engine. This is to encourage people to stay updated.

Finally, there is a new generic extraction API which is meant for use with the QGIS plugin. In particular, the following sub-APIs have been implemented to aggregate losses, damages and curves in the scenario risk, damage and event based calculators respectively:


If no tags are given, then full aggregation on all assets is performed. Using directly this API is discouraged, you should use the QGIS plugin instead, which has a nice GUI to define the aggregation queries. The API return .npz files, which are a good format for Python arrays.

oq commands#

The command oq info now understands an --extract flag that lists all the available extraction procedures.

We added a command oq plot_assets to plot all the assets involved in a risk calculation, together with the hazard sites.

The command oq plot_sites to plot hazard sites together with the hazard sources has been fixed, it had stopped working properly. Now it is automatically tested.

The command oq upgrade_nrml now understands a --multipoint flag which is useful to convert a source model containing pointSources into a model containing multiPointSources.

oq run was not logging properly in absence of rtree; this has been fixed.

oq reset and oq restore have been made more robust against corner cases.


From the IT point of view, the biggest change is the introduction of a new experimental distribution mechanism based on ZeroMQ. It can be enabled by setting oq_distribute=zmq in the configuration file openquake.cfg or by setting the environment variable OQ_DISTRIBUTE=zmq. The new distribution mechanism is not ready for production yet, but it has a huge potential for the future. On a single machine the zmq distribution has the following advantages, compared to the approach we are using now:

  1. it requires less memory (no fork)

  2. it allows to run much larger calculations (no more multiprocessing errors)

  3. it is a real queue, i.e. it manages multiple users well.

On a cluster, all of the above advantages apply and in addition we can save huge amount of memory now used by rabbitmq. Moreover, the engine will become a lot more HPC-friendly once rabbitmq and celery are abandoned.

Users interested in these kinds of features should contact us. For the moment, development on this front is slow since it is a low priority.


The classical_risk calculator can read the hazard curves from an XML file or from a CSV file in a custom format. The support for reading hazard curves in a CSV format has been deprecated since the format may be removed (we could read from an .hdf5 file instead).

The following exporters have been deprecated: hcurves-rlzs, agg_loss_table, losses_total, dmg_by_tag, dmg_total, losses_by_tag, losses_by_tag-rlzs, asset_loss_table.


A substantial amount of work went into testing and packaging. Now we have automated tests for macOS both for Python 2.7 and Python 3.5. Moreover celery is now tested on Linux both with Python 2.7 and Python 3.5. The Hazard Modeller’s Toolkit (HMTK) has been brought under continuous integration (CI).

As usual, the internal format in the datastore has changed, so you cannot read calculations generated by previous versions of the engine. Now the name of the Python processes spawned by the engine is oq-worker: this is convenient, both for visualization/listing purposes and also for killing all the engine processes at once.

We fixed an error in Windows caused by a random seed being a numpy.uint32 integer instead of a Python integer.

Our roadmap for abandoning Python 2 has been updated. In short, we will not abandon it until the QGIS plugin is ported to Python 3 and therefore we are waiting for QGIS 3.0 to become stable.