CBIIT / ccdi-dcc-federation-service / 29345067299 / 1
88%
main: 87%

LAST BUILD BRANCH: 1.2.0-test
DEFAULT BRANCH: main
Ran
Files 146
Run time 12s
Badge
Embed ▾
README BADGES
x

If you need to use a raster PNG badge, change the '.svg' to '.png' in the link

Markdown

Textile

RDoc

HTML

Rst
coverage: 87.849% (+0.1%) from 87.735%
29345067299.1

push

github

chaozhang-nci 
Dedupe the study set in the last two count queries that skipped it

`st2_list + st1_list` is list CONCATENATION, not union. A sample reaching one
study via BOTH the cell_line path and the participant path contributes that study
twice, so UNWINDing the list yields two identical rows for a single
(sample_id, study_id) pair. Queries that collapse those rows -- with WITH DISTINCT,
or a collect() grouped by the pair -- are unaffected. Two did not:

- the sample-node values query (tissue_type, tumor_classification,
  age_at_collection) counted raw rows, so a both-path sample would have its value
  counted twice while total counted the pair once (values_sum > total);
- _count_samples_by_diagnosis_category's total counted raw rows while its own
  missing and values queries count pairs, so it would report a different total
  than every other field for the same population.

Both now dedupe to the pair. An audit of every query that UNWINDs the study set
confirms these were the last two; all others already collapsed the duplicates.

Worth noting the invariant check could not have caught the first one: it bounds
values_sum from below (every non-missing pair contributes >= 1 value), so an
over-count satisfies it. No upper bound is possible, since multi-value fields
legitimately exceed total. It had to be fixed rather than monitored.

Behaviour-neutral on current data -- no sample has both a cell_line and a
participant, so all 15 countable fields stay byte-identical to the baseline.
Simulated against a both-path sample: before 2, after 1, for both queries.

25399 of 28912 relevant lines covered (87.85%)

0.88 hits per line

Source Files on job 29345067299.1