The root cause of the breakage appears to be that the hazard curve's
'statistic_type' is NULL but the risk code in question
(ClassicalPSHABasedMixin._get_db_curve()) expects it to be set to 'mean'.
Maybe the mean hazard curve is not being written to the database?
This is from openquake/risk/job/classical_psha.py, line 72
hc = models.HazardCurveData.objects.filter(
hazard_curve__output__oq_job=job,
hazard_curve__statistic_type='mean').extra(
where=["ST_GeoHash(location, 12) = %s"], params=[gh]).get()
I checked the geohash and it seems to be correct.
openquake=# SELECT * FROM hzrdr.hazard_curve WHERE output_id=509
id | output_id | end_branch_label | statistic_type | quantile
-----+-----------+------------------+----------------+----------
215 | 509 | 0 | |
openquake=# SELECT * FROM hzrdr.hazard_curve_data WHERE hazard_curve_id=215 AND ST_GeoHash(location, 12)='sqg1c4mr6jqf'
id | hazard_curve_id | poes | location
------+-----------------+-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+----------------------------------------------------
9369 | 215 | {1,0.999999999999453,0.999999999548169,0.999999836608758,0.99997616150681,0.998842529677148,0.981856818337621,0.889990232703232,0.68045118835525,0.429341652777471,0.226848426506306,0.106924265907651,0.0451297161821955,0.0167622312848525,0.00520462897461305,0.00126037470418106,0.000214021103311146,1.58934217476814e-05,0} | 0101000020E61000000AD7A3703D0A2F4085EB51B81E254340
Here's the problem:
A few weeks ago, mean hazard curves were always being computed (even if the param COMPUTE_
This is what caused the bug to surface.
When the Classical Risk calculator computes loss ratio curves for a given asset + location, a hazard curve is required to do the calculation. The hazard curve used should be the mean curve of the hazard curve computed for the site of interest over N logic tree samples. The loss ratio curve calculation function queries the database for a mean hazard at the given site.
The classical_
Here's what we need to do to fix it:
1) For Classical Risk calculations, COMPUTE_
2) We need appropriate database and job config constraints to enforce this rule. The rule is: if calculation mode is "classical" and job type is Hazard + Risk, then COMPUTE_
3) In the database, we are recording the 'job_type' in the oq_params table. This is really 'calc_mode' and should be renamed as such.
4) Once the existing 'job_type' is renamed to 'calc_mode', a new 'job_type' parameter should be introduced to record the type of the job, which at present is 'hazard' and/or 'risk'. This field should be a VARCHAR[] and should <= ['hazard', 'risk'] (one and/or the other).
5) Make any changes necessary to job config validation. I can't think of anything, but we may have to change something.
6) Finally, write a QA test that exercises this demo and validates the computed risk and hazard artifacts (curves, map, etc.).