| Age | Commit message (Collapse) | Author |
|---|
|
|
|
And remove n+1 queries
|
|
|
|
af83/6884-tomtom-matrix--handle-error-when-response-doesn,t-inclu
Handle API error(s) in Tomtom matrix. Fixes #6884
|
|
|
|
|
|
I previously tried to correct a circular dependency problem in
a057276129b1f62b811743db3b8f867a05241ed3, but that didn't fix it (it was
intermittent, and came back).
After some wrangling, I've now deduced with some confidence that the
problem comes from `RouteWayCostCalculator`, which used
`TomTom::Matrix::RemoteError`. From the way it looks, this seems to mess
up the Rails autoloader since `tom_tom.rb` will try to load the `Matrix`
class from the `TomTom.matrix` call above. Or something.
In an attempt to fix the circular dependency error for real this time,
move the error class to a completely separate module from `Matrix`, and
refer to this when we need to use the error class.
Refs #6884
|
|
If there's an API error, we shouldn't update the route's `costs` field.
Let's say we've already calculated some costs for a route A. We then edit
and re-save A, which triggers a recalculation of the costs. Now the
TomTom API responds with an error. We don't want to overwrite our
existing costs with an empty array because they could still be useful.
In this case, we should instead keep the existing costs we already had.
To achieve this, move the `RemoteError` rescue into
`RouteWayCostCalculator`, leaving the error unhandled in
`TomTom.matrix`.
Refs #6884
|
|
|
|
|
|
Calculate routes using the TomTom matrix API. Update the spec's HTTP
request stub accordingly (note that the stub is still fake).
Refs #6222
|
|
This new method will launch a worker that takes the route's `StopArea`s,
converts them to `WayCost`s, sends them to the TomTom API to calculate
distance and time costs, and saves those costs to the route's `costs`
field.
Refs #6095
|
|
|
|
Replace `ParentImportNotifier` with the new generalised
`ParentNotifier`. This will allow us to use the same service for parent
notification of both imports and compliance check sets.
Delete the `ParentImportNotifier` class and spec as these are now
superseded by `ParentNotifier`.
Refs #4758
|
|
This is a takeoff on `ParentImportNotifier`. I just copied over the
class and spec and generalised the service to work with `Import`-style
interfaces. We can now use the same notifier service class for both
imports and compliance check sets.
Refs #4758
|
|
|
|
|
|
|
|
Still missing:
- Pagination
- Filters
|
|
- Renaming as requested STIF::NetexFile::Frame.get_line_object_id -> get_sgort_id
- working on error_messages for imports#show
|
|
- Rebased #5006 upoon master and created #5281 from #5006
- Setup regression test based on client provided input
- Fixed bug with multiple spurious directory occurances
|
|
|
|
Implementing allowed vs foreign line lookup for the zip service
- Adapting `lib/stif/netex_file.rb` to expose the already implemented matching
- `app/services/zip_service.rb` augmented to check for allowed lines and returning forbidden lines
- Specs with fixtures and using the beforementioned new zip support in the specs
- Fixture directories for the new specs
|
|
|
|
|
|
Speced and implemented ZipService to return additional directories
in the entry's `spurious` field.
|
|
This is no longer being used as the `WorkbenchImportWorker` was
refactored and it was decided that we don't want to enable retries for
imports.
For example, imagine you have an hour-long import that fails after 50
minutes. We retry three times and now we've wasted 2.5 hours of your
time just to tell you that the import failed.
Thus the decision was to remove it.
|
|
Retry removed (but not the deadcode yet) [amend me]
|
|
Debugging Java Integration (ZipService -> HTTPService -> Object Creation)
- Created an ErrorFormat concern to see what went wrong in the HTTP call
as so many things can go wrong due to:
* Timeout Issues
* Illegal database setup or cleaning before tests
|
|
- Replzed Lazy Enum over get_next_entry and group by with explicit looping and yielding to an instance level enumerator
that yields streams per subdir (wich represents a referential).
- Fixtures and Specs that use a REAL usecase.
Next: Get rid of metadata kludge (should fix failing specs)
|
|
|
|
|
|
Instead of manually defining the finished statuses, use the method
that's already defined on `Import` to get them.
The only unfortunate part is that we now had to muss up our `Import`
code and convert those methods to class methods in order for them to be
accessible to `ParentImportNotifier` as well as its instance methods.
Refs #3511
|
|
This call can be written much more concisely. Thanks for the suggestion
Robert!
Refs #3511
|
|
Robert pointed out that it isn't necessary to prefix the method call
with `self.`. Makes the code cleaner.
Refs #3511
|
|
merge to keep a reference in feature branch)
|
|
A new service, `ParentImportNotifier`, will be called by a Cron job
every 5 minutes. This service will query all finished sub-imports that
haven't yet notified their parent, and tell them to do so.
factories/workbench_imports.rb:
Add a `WorkbenchImport` factory. This is currently incomplete. We'll
want to correctly initialise the `file` attribute eventually.
import.rb:
factories/imports.rb:
In order to get the test setup to work correctly, we skip the
`before_create` callback on `Import`. This is because that callback
initialises `status` to "new". We want to be able to set any status
value in our tests to test things properly. Thus the callback messes
with our buttermilk. For all imports created by the factory, skip this
callback. In order to skip the callback, I had to put it in its own
method and pass the method name to `#skip_callback`. That's why the
callback is now in a new method in `Import`.
Adds an empty `#notify_parent` method in `Import`. For now this doesn't
do anything, but eventually it will be filled in to perform the actual
notification. I this changeset, all we're doing is setting up the
connecting logic to notify the right things from the Cron job.
import_spec.rb:
Remove the `.class_name` validation because we removed that previously
in `Import`. The `belongs_to` is now a polymorphic association, with no
class name specified.
parent_import_notifier_spec.rb:
Needed to use `create` instead of `build_stubbed` in a lot of these
cases so that I could validate the query. Really don't like this because
now this set of tests is SUPER SLOW. If you have any ideas for how to
speed it up, let's do that.
Refs #3511
|
|
- Added tests for HTTPService (regression)
- Removed some dead code (Workbench model spec, Import factory)
- Changed front_end_host to rails_host in config
|
|
|
|
|
|
|
|
logging)
|
|
|
|
- speced but missing error treatment
|
|
|
