Event Timeline
Note: We will be re-pulling Android updated data after ~7 days post prod using new release data only (50428). We saw some negative results in data with prior 2 versions which should be fixed in new beta.
Updated data from v. 2.7.50428. See MAX values pivot for anomalies (events > 20000) by wiki/country.
@SNowick_WMF Could we look at average latencies daily, and grouped by wiki? Specifically we'd like to see if there have been any significant changes in latency for ptwiki in the last few days, since there have been changes in the restbase configuration which might impact latency.
(Note also that the latest version of the app (50431, released Feb 22) uses the new MEP session schema instead of the legacy one)
Hi @Dbrant looking at the schema doc for android-app-session (https://schema.wikimedia.org/repositories/secondary/jsonschema/analytics/mobile_apps/android_app_session/current.yaml) there is a column listed for page_load_latency_ms but I get 'can't seem to resolve the column "page_load_latency_ms"' errors querying for that data (and column doesn't show up when viewing schema from Hue). There is data in the length_ms column but that's a measurement for the entire session.
@Sharvaniharan might know more about why that column is not queryable...
In the meantime I can get the data from Legacy schema if you'd still like to see a date compare, lmk.
I believe you should use the parameter specified as session_data.page_load_latency_ms instead of just page_load_latency_ms.
In the meantime I can get the data from Legacy schema if you'd still like to see a date compare, lmk.
Sure, that would still be useful, if there isn't enough data in the new schema. The legacy schema should still be getting data from the tail end of the previous version of the app (version 50428) which haven't yet updated to the latest. But the sample sizes in the old schema are probably shrinking rapidly.
The following spreadsheet has query results for Android and iOS latency data for ptwiki as available given the constraints described above.
Looking for an increase after 2023-03-09 there does not seem to be any dramatic jumps on any of the datasets. The results with fewer unique users should be considered less reliable since we are working with averages and they are easily distorted by a few anomalous results.