SuccessChanges

Summary

  1. Handle two kinds of data.ts correctly (commit: eca6c2f3a7a945aa26573ed2e17275ddbfddb281) (details)
  2. Make the landing page responsive (commit: a3fde2c2685fc72cee93b1d55f11188da4192b17) (details)
  3. Switch the gzip open mode to be compatible with python3 (commit: 28bd097330ced145c7b7074b862dadb835a5d056) (details)
  4. Switch to reading the list of all users from the uuid_db (commit: f8cc80f1df36528a594f13992b9bb78bd5e1d737) (details)
  5. Fixes to firebase that are not tested by the test case (commit: 18769b8abddc9ca2c552d4d6a192f9780068ed1f) (details)
  6. Handle the case (on iOS) when the transition is sometimes None (commit: ac5ae40905ae188580158823e72af55bb9aa57b2) (details)
  7. Handle the case in which the user has only one place (commit: 0efdc1b5bdfec86928cccc50ffaa3292c6389e73) (details)
  8. Add a config option to turn off the filter_accuracy step without (commit: 6f85ff9f47ff22ffc3b02de308ce7dd86e9a48af) (details)
  9. Fix the issue with the output_gen state not marked as done (commit: 54c2dbc6c5ec02e644025fcf9149bf6aeaadcf59) (details)
Commit eca6c2f3a7a945aa26573ed2e17275ddbfddb281 by shankari
Handle two kinds of data.ts correctly
both strings (which are in ms) and floats (which are in seconds)
(commit: eca6c2f3a7a945aa26573ed2e17275ddbfddb281)
The file was modifiedemission/net/api/stats.py (diff)
Commit a3fde2c2685fc72cee93b1d55f11188da4192b17 by shankari
Make the landing page responsive
So that it is more clearly visible on small phone screens.
(commit: a3fde2c2685fc72cee93b1d55f11188da4192b17)
The file was modifiedwebapp/www/templates/client_setup.html (diff)
The file was modifiedwebapp/www/css/redirect.css (diff)
Commit 28bd097330ced145c7b7074b862dadb835a5d056 by shankari
Switch the gzip open mode to be compatible with python3
In python3, you need to open gzip files in text mode to be able to write
text to them. Otherwise, you need to convert to binary before writing. I
bet this is the outcome of the text changes
(commit: 28bd097330ced145c7b7074b862dadb835a5d056)
The file was modifiedbin/debug/extract_timeline_for_day_range_and_user.py (diff)
Commit f8cc80f1df36528a594f13992b9bb78bd5e1d737 by shankari
Switch to reading the list of all users from the uuid_db
Rationale:
- queries on timeseries are very slow
- we have entries in the timeseries, from stats conversion, that still
have
uuids from before the switch. We're keeping them around so that we can
do
analysis on the client timeseries alone, separated from
This implies that using the entries from the timeseries is not only slow
but incorrect. Let's just switching to reading entries from the uuid db.
We've had the client with the registration (to create uuid) forever -
anything older is probably incompatible anyway.
Let's also switch to the profile_db for platform specific queries. In
this case, the profile update was released months ago, so again, we
expect that all current clients should have updated by now.
(commit: f8cc80f1df36528a594f13992b9bb78bd5e1d737)
The file was modifiedbin/reset_pipeline.py (diff)
Commit 18769b8abddc9ca2c552d4d6a192f9780068ed1f by shankari
Fixes to firebase that are not tested by the test case
Unfortunately, it is hard to test this without actually sending messages
to firebase. Found while trying to do a live demo of push notifications.
We should really look up mocking to figure out how to test this.
(commit: 18769b8abddc9ca2c552d4d6a192f9780068ed1f)
The file was modifiedemission/net/ext_service/push/notify_interface_impl/firebase.py (diff)
Commit ac5ae40905ae188580158823e72af55bb9aa57b2 by shankari
Handle the case (on iOS) when the transition is sometimes None
let's just map None -> None. otherwise we end up clogging the logs with
error messages, and clogging the error_db with them as well
(commit: ac5ae40905ae188580158823e72af55bb9aa57b2)
The file was modifiedemission/net/usercache/formatters/ios/transition.py (diff)
Commit 0efdc1b5bdfec86928cccc50ffaa3292c6389e73 by shankari
Handle the case in which the user has only one place
We had already handled the case in which the user had no places. This
also handles the case in which the user has exactly one place. We ran
into this while resetting the pipeline while re-running after moving to
the more scalable server.
In that case too, we return None and skip the reset.
(commit: 0efdc1b5bdfec86928cccc50ffaa3292c6389e73)
The file was modifiedemission/pipeline/reset.py (diff)
The file was modifiedemission/storage/decorations/place_queries.py (diff)
Commit 6f85ff9f47ff22ffc3b02de308ce7dd86e9a48af by shankari
Add a config option to turn off the filter_accuracy step without
modifying code
As part of the change, restructured the analysis config into a separate
config object and changed the only existing analysis config usage to use
the new structure.
(commit: 6f85ff9f47ff22ffc3b02de308ce7dd86e9a48af)
The file was modifiedemission/analysis/intake/cleaning/clean_and_resample.py (diff)
The file was modifiedconf/analysis/debug.conf.json.sample (diff)
The file was modifiedemission/analysis/intake/cleaning/filter_accuracy.py (diff)
Commit 54c2dbc6c5ec02e644025fcf9149bf6aeaadcf59 by shankari
Fix the issue with the output_gen state not marked as done
We should now be able to turn on a cronjob to periodically run the
analysis pipeline
(commit: 54c2dbc6c5ec02e644025fcf9149bf6aeaadcf59)
The file was modifiedemission/tests/pipelineTests/TestPipelineReset.py (diff)
The file was modifiedemission/net/usercache/builtin_usercache_handler.py (diff)