* Configure ingest repo access/pool size
If I'm not mistaken 3 is a sane default, the only
inserts we're doing are:
- session buffer dump
- events buffer dump
- GA import dump
And all are serializable within their scopes?
* Add IngestRepo
* Start IngestRepo
* Use IngestRepo for inserts
* Annotate ClickhouseRepo as read_only
So no insert* functions are expanded
* Update moduledoc
* rename alias
* Fix default env var value so it can be casted
* Use IngestRepo for migrations
* Set default ingest pool size from 3 to 5
in case conns are restarting or else...
* Ensure all Repo prometheus metrics are collected
* Move Endpoint errors setup to common config
* Implement naive Sentry link resolver
* Implement error report e-mail
* Delete static sentry script
* Implement user feedback form on server errors
* Re-arrange pipe
* Use Sentry.Config.dsn() where applicable
* Fix typo
* Use Map.replace/3
This PR replaces geolix with locus to simplify self-hosted setup. locus can auto-update maxmind dbs which are recommended for self-hosters if they want city-level geolocation. locus is also a bit faster.
This PR also uses a test mmdb file from https://github.com/maxmind/MaxMind-DB for e2e geolocation tests without stubs.
* Ignore XX and T1 countries
* Add fallback if country_code=nil
* Lookup city overrides directly in CityOverrides module
* Changelog
* Add empty moduledoc
* Remove redundant comment
* Set pg pool size for MIX_ENV=test
* Include slow tests in CI run
* Exclude slow tests by default
* Mark tests slow/async where applicable
* Restructure captcha mocks
* Revert async where env is relied upon
* Add --max-failures=1 to CI run
* Set warnings as errors
* Disable async where various mocks are used
* Revert "Disable async where various mocks are used"
This reverts commit 2446b72a29.
* Disable async for test using vcr
### Changes
This PR adds a fallback to empty build metadata when BUILD_METADATA
contains invalid JSON.
Example `warning` log for `BUILD_METADATA={...}`:
```
20:57:57.872 [warning] failed to parse $BUILD_METADATA, reason: ** (Jason.DecodeError) unexpected byte at position 1: 0x2E (".")
```
Fixes https://github.com/plausible/analytics/issues/2491
### Tests
- [x] This PR does not require tests
### Changelog
- [ ] Entry has been added to changelog
### Documentation
- [x] This change does not need a documentation update
### Dark mode
- [x] This PR does not change the UI
* Update Sites.Cache
So it's now capable of refreshing most recent sites.
Refreshing a single site is no longer wanted.
* Introduce Warmer.RecentlyUpdated
This is Sites Cache warmer that runs only for
most recently updated sites every 30s.
* Validate Request creation early
* Rename RateLimiter to GateKeeper and introduce detailed policies
* Update events API tests - a provisioned site is now required
* Update events ingestion tests
* Make limits visible in CRM Sites index
* Hard-deprecate DOMAIN_BLACKLIST
* Remove unnecessary clause
* Fix typo
* Explicitly delegate Warmer.All
* GateKeeper.allwoance => GateKeeper.check
* Instrument Sites.Cache measurments
* Update send_pageview task to output response headers
* Instrument ingestion pipeline
* Credo
* Make event telemetry test a sync case
* Simplify Request.uri/hostname handling
* Use embedded schema, apply action and rely on get_field
* Implement sites by domain caching interface + warmer
* Add test
* Implement hit rate interface
* Add moduledocs
* Fix up typespec
* s/warmer/warmer_fn
* Extract measure_duration/2
* Fix up typespec
* Log errors and return nil on cache internal errors
* Fix up non-existing cache test
* Retrieve specific db columns when pre-filling the cache
* Reduce the subset of fields retrieved from the DB
See 63f3c6233d (r89871536)
* Implement FF-driven DB lookup for sites during ingestion
We like to see the impact of doing a simple postgres lookup on each
ingestion event. The percentage-based feature flag `:ingestion_pg_lookup`
must be set in order for lookups to be executed.
* Fix resolving Cachex stats metrics
* Enable PromEx on dev env
This pull request improves the current OpenTelemetry implementation. Currently only 1% of the spans are sent, due to the high volume of ingestion requests to /api/event. I enabled the 1% sampling to /api/event only, recording 100% of the other traces.
* remove tracker files from git index
* generate tracker files on npm test
* generate tracker files for elixir tests/dev/CI
* update tracker/package-lock.json
* exclude npm run deploy from mix test + some docs
Getting this error when running the release:
ERROR! the application :fun_with_flags has a different value set for key :persistence during runtime compared to compile time. Since this application environment entry was marked as compile time, this difference can lead to different behaviour than expected:
* Compile time value was not set
* Runtime value was set to: [adapter: FunWithFlags.Store.Persistent.Ecto, repo: Plausible.Repo]
* List all Google Analytics views during import
This commit fixes a bug where different Google Analytics views with the
same name and URI were not shown. This was caused because GA views were
stored as a map, that naturally doesn't support duplicate keys.
This change updates the GA views list to display view IDs, making it
clearer to know what is being imported. The dropdown is now grouped by
website URL.
* Put Google Analytics API URLs in app env
* Add controller test to GA view list
* Remove invalid Jason.decode argument
Co-authored-by: Robert Joonas <robertjoonas16@gmail.com>
* Add custom message to Google invalid grant error
Co-authored-by: Robert Joonas <robertjoonas16@gmail.com>
* Test invalid_grant while refreshing Google token
Co-authored-by: Robert Joonas <robertjoonas16@gmail.com>
Co-authored-by: Robert Joonas <robertjoonas16@gmail.com>
* Create separate module for GA HTTP requests
* Fetch GA data entirely instead of monthly
* Add buffering to GA imports
* Change positional args to maps when serializing from GA
* Create Google Analytics VCR tests
* Fix geolocation subdivision pattern matching
This commit fixes a bug where regions were not being saved. This was
caused because Geolix response was returning an additional
`:geolocation` map key. It also adds a test case for this.
Closes#2033
* Add geolocation database to .gitignore
* Upgrade geolix
* Remove geolix pool config
* Save unnecessary Task.async_stream roundtrip
Normally the Geolix API accepts `:where` keyword option that designates
the database to look up. In case no parameter is supplied, it'll spawn
a parallel map over all databases available. In this case we have only
one DB anyway, so there is no need for the extra instrumentation.
* Follow up on direct :geolocation lookups
* Introduce Finch for Sentry integration
* Make sure the DummyAgent can be started
* No need to sanitize the dsn, finch takes care of that
* Simplify the dummy child spec
* Annotate redirects clause
* Make use of new `get_int_from_path_or_env`
* Actually use finch in Sentry config
* Configure `excluded_domains` correctly for Sentry
The way sentry is configured currently, when we get an HTTP error it
will be logged twice - once from Sentry.PlugCapture and once from
Sentry.LoggerBackend. The logger backend module does the right thing
by default but for some reason we've been overriding the config
parameter that by default stops double-counting errors. This commit
returns to the default configuration which is better.
* Default to 15s timeout
* Attempt to send twice at most
* Warn in sentry client
* Use warn level in sentry client
Co-authored-by: Adam Rutkowski <hq@mtod.org>
* Include gelocation DB download in the development workflow
* Make sure `tls_certificate_check` is started ASAP
This prevents `:application_either_not_started_or_not_ready` errors
on application startup.
* Mark Makefile targets as PHONY
By default Make assumes the targets are files,
in this case none of them are.
* Adds tri-state disable_registration config
* Formatting
* Changes variable back to atom
* Changelog
* Uses atoms correctly :/
* Swaps to a more fitting value
* Formatting
* Add has_imported_stats boolean to Site
* Add Google Analytics import panel to general settings
* Get GA profiles to display in import settings panel
* Add import_from_google method as entrypoint to import data
* Add imported_visitors table
* Remove conflicting code from migration
* Import visitors data into clickhouse database
* Pass another dataset to main graph for rendering in red
This adds another entry to the JSON data returned via the main graph API
called `imported_plot`, which is similar to `plot` in form but will be
completed with previously imported data. Currently it simply returns
the values from `plot` / 2. The data is rendered in the main graph in
red without fill, and without an indicator for the present. Rationale:
imported data will not continue to grow so there is no projection
forward, only backwards.
* Hook imported GA data to dashboard timeseries plot
* Add settings option to forget imported data
* Import sources from google analytics
* Merge imported sources when queried
* Merge imported source data native data when querying sources
* Start converting metrics to atoms so they can be subqueried
This changes "visitors" and in some places "sources" to atoms. This does
not change the behaviour of the functions - the tests all pass unchanged
following this commit. This is necessary as joining subqueries requires
that the keys in `select` statements be atoms and not strings.
* Convery GA (direct) source to empty string
* Import utm campaign and utm medium from GA
* format
* Import all data types from GA into new tables
* Handle large amounts of more data more safely
* Fix some mistakes in tables
* Make GA requests in chunks of 5 queries
* Only display imported timeseries when there is no filter
* Correctly show last 30 minutes timeseries when 'realtime'
* Add with_imported key to Query struct
* Account for injected :is_not filter on sources from dashboard
* Also add tentative imported_utm_sources table
This needs a bit more work on the google import side, as GA do not
report sources and utm sources as distinct things.
* Return imported data to dashboard for rest of Sources panel
This extends the merge_imported function definition for sources to
utm_sources, utm_mediums and utm_campaigns too. This appears to be
working on the DB side but something is incomplete on the client side.
* Clear imported stats from all tables when requested
* Merge entry pages and exit pages from imported data into unfiltered dashboard view
This requires converting the `"visits"` and `"visit_duration"` metrics
to atoms so that they can be used in ecto subqueries.
* Display imported devices, browsers and OSs on dashboard
* Display imported country data on dashboard
* Add more metrics to entries/exits for modals
* make sure data is returned via API with correct keys
* Import regions and cities from GA
* Capitalize device upon import to match native data
* Leave query limits/offsets until after possibly joining with imported data
* Also import timeOnPage and pageviews for pages from GA
* imported_countries -> imported_locations
* Get timeOnPage and pageviews for pages from GA
These are needed for the pages modal, and for calculating exit rates for
exit pages.
* Add indicator to dashboard when imported data is being used
* Don't show imported data as separately line on main graph
* "bounce_rate" -> :bounce_rate, so it works in subqueries
* Drop imported browser and OS versions
These are not needed.
* Toggle displaying imported data by clicking indicator
* Parse referrers with RefInspector
- Use 'ga:fullReferrer' instead of 'ga:source'. This provides the actual
referrer host + path, whereas 'ga:source' includes utm_mediums and
other values when relevant.
- 'ga:fullReferror' does however include search engine names directly,
so they are manually checked for as RefInspector won't pick up on
these.
* Keep imported data indicator on dashboard and strikethrough when hidden
* Add unlink google button to import panel
* Rename some GA browsers and OSes to plausible versions
* Get main top pages and exit pages panels working correctly with imported data
* mix format
* Fetch time_on_pages for imported data when needed
* entry pages need to fetch bounces from GA
* "sample_percent" -> :sample_percent as only atoms can be used in subqueries
* Calculate bounce_rate for joined native and imported data for top pages modal
* Flip some query bindings around to be less misleading
* Fixup entry page modal visit durations
* mix format
* Fetch bounces and visit_duration for sources from GA
* add more source metrics used for data in modals
* Make sources modals display correct values
* imported_visitors: bounce_rate -> bounces, avg_visit_duration -> visit_duration
* Merge imported data into aggregate stats
* Reformat top graph side icons
* Ensure sample_percent is yielded from aggregate data
* filter event_props should be strings
* Hide imported data from frontend when using filter
* Fix existing tests
* fix tests
* Fix imported indicator appearing when filtering
* comma needed, lost when rebasing
* Import utm_terms and utm_content from GA
* Merge imported utm_term and utm_content
* Rename imported Countries data as Locations
* Set imported city schema field to int
* Remove utm_terms and utm_content when clearing imported
* Clean locations import from Google Analytics
- Country and region should be set to "" when GA provides "(not set)"
- City should be set to 0 for "unknown", as we cannot reliably import
city data from GA.
* Display imported region and city in dashboard
* os -> operating_system in some parts of code
The inconsistency of using os in some places and operating_system in
others causes trouble with subqueries and joins for the native and
imported data, which would require additional logic to account for. The
simplest solution is the just use a consistent word for all uses. This
doesn't make any user-facing or database changes.
* to_atom -> to_existing_atom
* format
* "events" metric -> :events
* ignore imported data when "events" in metrics
* update "bounce_rate"
* atomise some more metrics from new city and region api
* atomise some more metrics for email handlers
* "conversion_rate" -> :conversion_rate during csv export
* Move imported data stats code to own module
* Move imported timeseries function to Stats.Imported
* Use Timex.parse to import dates from GA
* has_imported_stats -> imported_source
* "time_on_page" -> :time_on_page
* Convert imported GA data to UTC
* Clean up GA request code a bit
There was some weird logic here with two separate lists that really
ought to be together, so this merges those.
* Fail sooner if GA timezone can't be identified
* Link imported tables to site by id
* imported_utm_content -> imported_utm_contents
* Imported GA from all of time
* Reorganise GA data fetch logic
- Fetch data from the start of time (2005)
- Check whether no data was fetched, and if so, inform user and don't
consider data to be imported.
* Clarify removal of "visits" data when it isn't in metrics
* Apply location filters from API
This makes it consistent with the sources etc which filter out 'Direct /
None' on the API side. These filters are used by both the native and
imported data handling code, which would otherwise both duplicate the
filters in their `where` clauses.
* Do not use changeset for setting site.imported_source
* Add all metrics to all dimensions
* Run GA import in the background
* Send email when GA import completes
* Add handler to insert imported data into tests and imported_browsers_factory
* Add remaining import data test factories
* Add imported location data to test
* Test main graph with imported data
* Add imported data to operating systems tests
* Add imported data to pages tests
* Add imported data to entry pages tests
* Add imported data to exit pages tests
* Add imported data to devices tests
* Add imported data to sources tests
* Add imported data to UTM tests
* Add new test module for the data import step
* Test import of sources GA data
* Test import of utm_mediums GA data
* Test import of utm_campaigns GA data
* Add tests for UTM terms
* Add tests for UTM contents
* Add test for importing pages and entry pages data from GA
* Add test for importing exit page data
* Fix module file name typo
* Add test for importing location data from GA
* Add test for importing devices data from GA
* Add test for importing browsers data from GA
* Add test for importing OS data from GA
* Paginate GA requests to download all data
* Bump clickhouse_ecto version
* Move RefInspector wrapper function into module
* Drop timezone transform on import
* Order imported by side_id then date
* More strings -> atoms
Also changes a conditional to be a bit nicer
* Remove parallelisation of data import
* Split sources and UTM sources from fetched GA data
GA has only a "source" dimension and no "UTM source" dimension. Instead
it returns these combined. The logic herein to tease these apart is:
1. "(direct)" -> it's a direct source
2. if the source is a domain -> it's a source
3. "google" -> it's from adwords; let's make this a UTM source "adwords"
4. else -> just a UTM source
* Keep prop names in queries as strings
* fix typo
* Fix import
* Insert data to clickhouse in batches
* Fix link when removing imported data
* Merge source tables
* Import hostname as well as pathname
* Record start and end time of imported data
* Track import progress
* Fix month interval with imported data
* Do not JOIN when imported date range has no overlap
* Fix time on page using exits
Co-authored-by: mcol <mcol@posteo.net>
* Merge branch 'plausible_master'
* Add City level details
* Add City level details
* Use ISO codes instead of geoname_id for subdivisions
* Add easier way to configure geolocation database
* Add workflow for dev branch
* Correct clickhouse migration
* Translate subdivision names
* Translate city names
* WIP
* Region and country filters
* Fix region filter
* Remove region_name when removing region filter
* Add modals for regions and cities
* Remove dead code
* WIP
* Revert "WIP"
This reverts commit 3202bf2fe9.
* Feature flag to hide cities when deployed
* Add changelog entry
* Remove unused code
* Remove unused variables
* Fix test
Co-authored-by: AymanTerra <aymanterra@yahoo.com>
Phoenix only requires SECRET_KEY_BASE to be longer than 32 characters. 32 characters even with hex encoding is 128 bits of entropy which is more than enough. Some secret generation tools only generate 32 characters/128 bits by default which makes this relatively arbitrary limit annoying. Let's change it to what Phoenix requires.
* Invite existing user to a site
* Add invitation flow for non-existing users
* Accept and reject invitations
* Use invitation flow for existing users
* Locking mechanism for sites
* Authorization for site settings
* Show usage based on site ownership
* Add ability to remove members from a site
* Do not show settings link to viewer roles
* Ability to remove invitations
* Remove `Plausible.Sites.count_for/1`
* Fix tests
* Do not show the trial banner after the trial
* Correct trial emails
* Transfer ownership
* Send invitation email to existing user
* Add invitation email flows
* Add plug for role-based authorization
* Rename AuthorizeStatsPlug -> AuthorizeSiteAccess
* Add email flow for ownership transfer
* Fix URLs in emails
* Fix small copy issues
* Make 'People' its own section in site settings
* Notify user via email if their access has been removed
* Check site lock status when invitation is accepted
* Check lock status when user subscribes
* Make sure only admins and owners can create shared links
* Changelog
* Add LockSites to daily cron
* Clean invitations after 48 hours
* Add notices about expiry
* Add invitation expired page
* Add doc link
Currently, both are hard-coded to relatively small values.
This may cause an unnecessary high load on the Clickhouse instance.
It makes sense to make those configurable.
