| Age | Commit message (Collapse) | Author |
|---|
|
|
|
|
|
Fix some issues caught by tests due to changes introducing the hand-off of the
partial correlations computations to an external process
Fix some issues due to the changes that introduce context managers for
database connections
Update some tests to take the above two changes into consideration
|
|
Use the `with` context manager to open database connections, so as to ensure
that those connections are closed once the call is completed. This hopefully
avoids the 'too many connections' error
|
|
|
|
* Document the things that the partial correlations feature depends on to work
|
|
Long-running computations are handed off to external processes. This avoids
timeouts in the webserver, and also reduces chances of instability of the
webserver.
The results of these long-running computations are needed eventually, so this
commit provides a way to check for the state of the computation, and the
results if any.
|
|
Since the worker polls a queue for jobs, it can get into a busy poll. This was
the reason that there was a delay of 0.1 seconds between each poll instance.
This commit takes this a little further by doing an incremental backoff, where
it waits longer and longer after each poll where it does not find a job
available, up to an arbitrary maximum.
|
|
Run the partial correlations code in an external python process decoupling it
from the server and making it asynchronous.
Summary of changes:
* gn3/api/correlation.py:
- Remove response processing code
- Queue partial corrs processing
- Create new endpoint to get results
* gn3/commands.py
- Compose the pcorrs command to be run in an external process
- Enable running of subprocess commands with list args
* gn3/responses/__init__.py: new module indicator file
* gn3/responses/pcorrs_responses.py: Hold response processing code extracted
from ~gn3.api.correlations.py~ file
* scripts/partial_correlations.py: CLI script to process the pcorrs
* sheepdog/worker.py:
- Add the *genenetwork3* path at the beginning of the ~sys.path~ list to
override any GN3 in the site-packages
- Add any environment variables to be set for the command to be run
|
|
correlations
In the original version of the if statement* I believe it was
interpreted as "if a_val and (b_val is not None)". This caused
values of 0 for a_val (the primary trait's values) to be evaluated as
False.
I changed it to compare both a_val and b_val to None. This seems to have
fixed the issue.
* if (a_val and b_val is not None)
|
|
Context managers should be preferred when allocating resources.
* gn3/computations/wgcna.py (stream_cmd_output): Call Popen with context
manager.
|
|
When the encoding is not specified explicitly, the system default encoding is
used. This is not recommended.
* gn3/computations/ctl.py (call_ctl_script),
gn3/computations/gemma.py (generate_pheno_txt_file),
gn3/computations/parsers.py (parse_genofile),
gn3/computations/partial_correlations.py (partial_correlations_fast),
gn3/computations/rqtl.py (process_rqtl_output, process_perm_output),
gn3/computations/wgcna.py (dump_wgcna_data, call_wgcna_script),
gn3/fs_helpers.py (jsonfile_to_dict): Explicitly specify UTF-8 to be the file
encoding.
*
tests/unit/computations/test_gemma.py (TestGemma.test_generate_pheno_txt_file),
tests/unit/computations/test_wgcna.py (TestWgcna.test_create_json_file): Test
for call to open with encoding='utf-8' argument.
|
|
|
|
* Use `with` in place of plain `open`
* Use f-strings in place of `str.format()`
* Remove string interpolation from queries - provide data as query parameters
* other minor fixes
|
|
* tests/unit/db/test_genotypes2.py: New file
|
|
Test that the partial correlations endpoint handles a mix of existing and
non-existing control traits gracefully and issues a warning to the user.
Summary of changes:
* gn3/computations/partial_correlations.py: Issue a warning for all
non-existing control traits
* gn3/db/partial_correlations.py: update queries - use `INNER JOIN` for tables
instead of comma-separated list of tables
* tests/integration/conftest.py: Add `db_conn` fixture to provide a database
connection to the tests. This will probably be changed in the future to
connect to a temporary database for tests.
* tests/integration/test_partial_correlations.py: Add test to check for
correct behaviour with a mix of existing and non-existing control traits
|
|
Test that if the endpoint is queried and not a single one of the control
traits exists in the database, then the endpoint will respond with a
404 (not-found) status code.
Summary of changes:
* gn3/computations/partial_correlations.py: Check whether any control trait is
found. If none is found, return "not-found" message.
* gn3/db/partial_correlations.py: Fix bug in Geno query.
* tests/integration/test_partial_correlations.py: Add test for non-existing
control traits. Rename function to make it clearer what it is testing
for. Remove obsoleted comments.
|
|
Related to commit 75dcfe295af57b16428c586cc11dbaa827a5feba
This commit removes the related test that was checking for the wrong thing.
|
|
The code was migrated from GN1 with a faulty assumption that all trait types
have a corresponding `*Freeze` table in the database. This assumption is not
true for the `Temp` traits.
This commit removes the buggy code.
|
|
Test that the partial correlations endpoint responds with an appropriate
"not-found" message and the corresponding 404 status code in the case where a
request is made and the primary trait requested for does not exist in the
database.
Summary of the changes in each file:
* gn3/api/correlation.py: generalise the building of the response
* gn3/computations/partial_correlations.py: return with a "not-found" if the
primary trait does not exist in the database
* gn3/db/partial_correlations.py: Fix a number of bugs that led to exceptions
in the case that the primary trait did not exist
* pytest.ini: register a `slow` pytest marker
* tests/integration/test_partial_correlations.py: Add a new test to check for
an appropriate 404 response in case of a primary trait that does not exist
in the database.
|
|
Add a test for the partial correlations endpoint, with:
- no data in the request
- missing items in the data
Fix the bugs caught by the test
|
|
Create a client fixture to help with the integration tests
|
|
Use pytest's `mark` feature to explicitly categorise the tests and run them
per category
|
|
Add property tests using pytest and hypothesis to test that the expected
properties hold for the
`gn3.computations.partial_correlations.dictify_by_samples`
function.
|
|
|
|
These tools have been greatly improved and moved to a new home at
https://git.genenetwork.org/arunisaac/dump-genenetwork-database
* sql/map-database.sh, sql/schema-from-in-db-documentation.org,
sql/schema-original.sql, sql/schema.png, sql/schema.sql, sql/schema.svg: New
files.
|
|
Enable running commands:
- `python3 setup.py unit_check`: run the unit tests
- `python3 setup.py integration_check`: run integration tests
- `python3 setup.py performance_check`: run performance tests
|
|
Quote the shell variables to prevent globbing and word splitting.
Deactivate this check for the specific lines that require intentional word
splitting
|
|
We have our own laminar CI and no longer need GitHub actions.
* .github/workflows: Delete directory.
|
|
|
|
Do all the work in a single iteration to avoid unnecessary iterations that
hamper performance.
|
|
Web servers are long-running processes, and python is not very good at
cleaning up after itself especially in forked processes - this leads to memory
errors in the web-server after a while.
This commit removes the use of multiprocessing to avoid such failures.
|
|
|
|
This commit refactors the code to make it possible to use multiprocessing to
speed up the computation of the partial correlations.
The major refactor is to move the `__compute_trait_info__` function to the
top-level of the module, and provide to it all the other necessary context via
the new args.
|
|
In Python3 when slicing,
seq[:min(some_val, len(seq))] == seq[:some_val]
because Python3 will just return a copy of the entire sequence if `some_val`
happens to be larger/greater than the length of the sequence.
This commit removes the unnecessary call to `min()`
|
|
If a user replaces an individual value with an "x", delete that date entry
from the respective table. Deletion here is the only option since by default
the Nstrain, PublishData and PublishSE don't accept null values. Note that
deleting all 3 values is equivalent to removing the sample from the CSV file.
* gn3/db/traits.py (update_sample_data): If a value is "x", delete it from the
respective table.
|
|
When editing values from "x" to "0"(or any other value) when editing data, an
"update" statement was being run; thereby no new value was being inserted. To
the end user, modifying an "x" value to something else meant that no value was
being inserted. This commit fixes that by doing an insert whenever a change
from "x" to "0" is performed.
* gn3/db/traits.py (update_sample_data): Add insert statements whenever an
"update" statement returns a 0 row-count.
|
|
|
|
|
|
The rqtl_wrapper script was throwing an error when only a single
categorical covariate was used. This is apparently because "covars[,name]"
throws an error in such a situation. Using just "covars" in such a
situation prevents the error. So I just added an if statement checking
the number of covariates. There might be some better way to deal with
this in R, but this is the best I could come up with.
|
|
The PublishFreeze table isn't necessary in phenotype queries, since
PublishFreeze.Id = InbredSet.Id (for the purposes of identifying traits,
at least)
|
|
In line 91 of gn3/db/traits.py, there was an if statement "if
record[key] else 'x'" that was treating values of 0 as False, so I
changed it to explicitly check that values aren't None
|
|
The PublishFreeeze table is actually unnecessary for this query, since
the group ID (inbred_set_id) should be passed in and that ID is in the
PublishXRef table (so no neeed to join with PublishFreeze)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
