* Ensure only complete imports are considered in site imports data migration
* Refactor `SiteImports` data migration for clarity (h/t @RobertJoonas)
* Fix tests
* Create a worker to clean clickhouse deleted sites data
The plan is to run this weekly, but going to trigger it manually the first few times on cloud
* Make asserting count more reliable
* credo
* PR feedback
* Fixes
* Update request logging
Ultimate goal is to be able to compare results with and without a flag against each other.
To do this we need logging which displays the full request url with parameters.
Example logs:
```
14:46:09.042 request_id=F8MRLSsaKB7BeIkAAAHk [info] (200) GET /api/sites took 17ms
14:46:09.175 request_id=F8MRLTKYV3G-GqEAAAZB [info] (200) GET /api/stats/dummy.site/current-visitors took 24ms
14:46:09.396 request_id=F8MRLUDfav28LIkAAAIE [info] (202) POST /api/event took 5ms
14:46:09.501 request_id=F8MRLUDS_YhftUkAAAAD [info] (200) GET /api/stats/dummy.site/sources?period=30d&date=2024-04-04&filters=%7B%7D&with_imported=true&limit=9 took 111ms
14:46:09.508 request_id=F8MRLUDhHbK8WKUAAAah [info] (200) GET /api/stats/dummy.site/main-graph?period=30d&date=2024-04-04&filters=%7B%7D&with_imported=true&metric=visitors took 117ms
14:46:09.511 request_id=F8MRLUDS1CYntK4AAAaB [info] (200) GET /api/stats/dummy.site/entry-pages?period=30d&date=2024-04-04&filters=%7B%7D&with_imported=true&limit=9 took 121ms
14:46:09.541 request_id=F8MRLTk5sIPYSn4AABoC [info] (200) GET /api/stats/dummy.site/top-stats?period=30d&date=2024-04-04&filters=%7B%7D&with_imported=true&comparison=previous_period&compare_from=undefined&compare_to=undefined&match_day_of_week=true took 278ms
```
* re-add plug
* router_dispatch -> endpoint
* Always scope import ID by site as well
* Do not schedule new import job if there are any site imports in progress
* Disable import buttons when any import is in progress
* Simplify `schedule_job/4` (h/t @RobertJoonas)
changing the copy here as I think that in some situations the "we're still counting stats" message is now shown even to those dashboards where we've stopped counting stats so best to avoid that
* Give a more semantic name to a function
* Make the LineGraph component thinner
* Move LineGraph into a separate file
* Move interval logic into interval-picker.js
This commit also fixes a bug where the interval name displayed inside
the picker component flickers the default interval when the graph is
loading.
The problem was that we were counting on graphData for returning us the
current interval: `let currentInterval = graphData?.interval`
We should always know the default interval before making the main-graph
request. Sending graphData to IntervalPicker component does not make
sense anyway.
* extract data fetching functions out of VisitorGraph component
* Return graph_metric key from Top Stats API
This commit introduces no behavioral changes - only starts returning an
additional field, allowing us to avoid the following logic in React:
1. Finding the metric names, given a stat display name. E.g.
`Unique visitors (last 30 min) -> visitors`
2. Checking if a metric is graphable or not
* Move metric state into localStorage
This commit gets rid of the internal `metric` state in the VisitorGraph
component and starts using localStorage for that instead.
This commit also chains the main-graph request into the top-stats request
callback - meaning that we'll always fetch new graph data after top stats
are updated. And we do it all in a single function.
Doing so simplifies the loading state significantly, and also helps to
make it clear, that at all times, existing top stats are required before
we can fetch the graph. That's because the metric is determined by which
Top stats are returned (for example, we can't be sure whether revenue
metrics will be returned or not).
* Make sure graph tooltip says "Converted Visitors"
* Extract a StatsExport function component
Again, instead of relying on `graphData?.interval` we can read it from
localStorage, or default to the largest interval available. The export
should not be dependant on the graph.
* Extract SamplingNotice function component
* Extract WithImportedSwitch function component
* Stop "lazy-loading" the graph and top stats
Since the container is always on top on the page, it will be visible on
the first render in any case - no matter the screen size.
* Turn VisitorGraph into a function component
* Display empty container until everything has loaded
* Do not display loading spinner on realtime ticks
* Turn Top Stats into a fn component
* fetch top stats and graph async
* Make sure revenue metrics can remain on the graph
* Add an extra check to canMetricBeGraphed
* fix typo
* remove redundant double negation
* Move experimental_session_count? logic to within query object
* WIP new querying system for deciding what tables to query
* both -> either
* Include sample_percent in both tables
* Remove a hanging TODO
* Allow filtering by visit props on event queries if flag is on
* Make default sessions join more conditional
* Simplify events_join_sessions?
* Add some TODOs
* Fix assignment
* Handle entry/exit page visit props separately from props stored in events table
* Update test which created sessions/events differently from everyone else
* Make query_events private
* Dont filter by session properties on events table if querying sessions and joining in events
* Handle visits, pageviews, events and visitors metrics from other table
* both -> either
* events, pageviews are strictly event metrics
* Add support for (plain) breakdowns deciding which table to use
* Run tests with experimental_reduced_joins as a separate job
Also refactor which tests are run with postgres:15 to reduce number of jobs
* moduledocs for TableDecider
* Fix matrix
* Custom build name
* Move TEST_EXPERIMENTAL_REDUCED_JOINS check
* Handle percentage separately from other metrics
* Remove debug code
* TableDecider tests
* both => sample_percent
* Improve naming
* Simplify code
* Breakdowns retain old behavior if getting metric visitors
* Unify behavior of entry/exit page hostnames with rest
* Fix test naming
* CH Migration: exit/entry hostnames in sessions_v2
* Leave only exit_page_hostname, we already record hostnames
* Use ClickHouse DDL in favour of ecto so that cluster is included
* Compress with ZSTD(3)
* Expose Hostname filter in the dashboard dropdown
* Add `exit_page_hostname` to ClickHouse `sessions_v2` schema
* Start tracking hostname changes in sessions
* Implement hostname filter suggestions
* Enable filtering by `event:hostname`
* Add tests for filtering by hostnames
* Ensure filter suggestions work for exit pages too
* Allow overriding hostnames with `send_pageview` mix task
* Remove `:window_time_on_page` flag
It seems that we can remove it after all?
* Initialize `experimental_hostname_filter` query parameter
* Rewrite cache store behaviour with regards to session hostnames
* Work around inconsistent session merging
So that `populate_stats` can get closer to actual ingestion
* Improve top stats test
* Make it possible to filter sessions by entry/exit hostnames
* Update pages tests
* Expose `experimental_hostname_filtering` temporarily in the UI
* Untested yet: also apply experimental filtering to sources
* Introduce `hostname_filter` feature flag
* Format
* Test top sources with hostname filter + experimental flag
* CH Migration: exit/entry hostnames in sessions_v2
* Leave only exit_page_hostname, we already record hostnames
* Use ClickHouse DDL in favour of ecto so that cluster is included
* Compress with ZSTD(3)
* Check import presence across all imports and not just the first one
Also, simplify imported data toggle rendering to not explicitly
refer to the earliest import source.
* Change imported stats toggle icon in dashboard
* Test `Imported.get_imports_date_range/1`
* Simplify failed UA/GA import email copy
* Always select and clear import ID 0 when referring to legacy imports
* Implement script for adding site import entries and adjusting end dates
* Log cases where end date computation is using fallback
* Don't log queries when running the migration to reduce noise
* Implement adjusting imported date range to actual and existing stats
* Drop redundant prefix from import list entries
* Make pageview numbers in imports list formatted for readability
* Test and improve date range cropping
* DRY UA and GA4 stats start and end date API calls
* Extend UA/GA import controller tests and improve error handling
* refactor finding longest open range without existing data
* Fix typo in test description
Co-authored-by: RobertJoonas <56999674+RobertJoonas@users.noreply.github.com>
* Rename `open_ranges` to `free_ranges`
---------
Co-authored-by: Robert Joonas <robertjoonas16@gmail.com>
Co-authored-by: RobertJoonas <56999674+RobertJoonas@users.noreply.github.com>
* Use sessionStorage for offer e-mail report banner tracking
Keeping it within the cookie is problematic, as the banners don't
expire and overflow the cookie with data when enough new sites
are added.
Ref https://github.com/plausible/analytics/issues/3762
* Update changelog
* Extract a component
* Make is_dbip evaluate to quoted boolean
* Ignore sessions without entry/exit pages when breaking down entry/exit pages
* Update stats controller tests to have more realistic test data (pageview followed by event)
* Add validation for the events metric in main_graph
* Test the already existing events metric support in main-graph
* Put total conversions on the graph
* extract main_graph_csv function (refactor only)
* add total_conversions and conversion_rate to goal-filtered visitors.csv
* update changelog
* add conversion rate to Stats API timeseries
* make sure CR can be queried as the only metric
* add a test asserting zeros are returned
* add tests for filtering by other properties at the same time
* Remove unnecessary validation of params
1. It doesn't make to validate `interval` (and its granularity) in all
endpoints. It's only relevant for the main graph.
2. The plug (renamed to `date_validation_plug`) already makes sure that
the dates are validated. No need to call the same function again in
Top Stats and Funnel endpoints.
* add metric validation to main graph
* Add tests for main graph API
* put conversion rate on the graph
* update changelog
* Add revenue metrics into metrics.ex
* make fn private
* avoid setting graph metric to visitors in goal-filtered view
* Unify GA4 and UA import flow into one
* Clean up property and view data retrieval via Google HTTP APIs
* Turn `Map.get` into `Map.fetch!` in API response processing code
* Bump list account summaries page size limit to max of 200
* Show only views in legacy flow and fix legacy redirect after import start
* Move google analytics import actions tests to a separate module
* Extend Google Analytics controller tests
* DRY up `property?` predicate (h/t @RobertJoonas)
* Revert "Revert "Move conversion_rate logic from elixir to clickhouse (#3887)"…"
This reverts commit 253fb5d67d.
* Fix issue with missing columns
The issue came from refactoring event:goal UNION ALL logic and trying to move
name select from first to last. If any other tables were joined, the incorrect
item would be used as an array index, causing this issue.
Added a relevant test.
* Separate out query building from pagination/execution logic.
* Refactor pageview_goals breakdown query, removing index column from results
* Remove zip_columns logic
* Use common pagination util
* Do everything in a single query for breakdowns for goals
* Order in DB
* Make sure column order is identical
* Calculate CR within the goal breakdown query
* Calculate CR for property breakdowns
* WIP: Calculate group CR
* CR with order_by
* Compatibility fix
* Import Ecto.Query and cleanup
* handle total_visitors the same way as add_percentage
* Handle conversion_rate in aggregate.ex
* Solve rebase fail
* Simplify maybe_add_group_conversion_rate
* Add conversion_rate defaults to 0 test
* Add test for conversion_rate should not be calculated with imported data (failing here and on master)
* Dont include imported data when breakdown by prop or goal
* Remove revenue_nils
