Use Blacksmith checkout with sticky-disk git mirror in Elixir CI - #6564

…t time) (plausible#6571)

* Revert "Use Blacksmith checkout with sticky-disk git mirror in Elixir CI (plausible#6564)"

This reverts commit fd528f2.

* Fix inert service container health checks in Elixir CI

The postgres and clickhouse health-check options were nested under env,
so Docker never received a health check and jobs could start before
ClickHouse was listening, failing ecto.create with
Mint.TransportError{reason: :closed}. Move options to the service level
and point the ClickHouse check at the real HTTP port (8123 /ping)
instead of the unused 8124.

Co-authored-by: Codesmith <codesmith-bot@users.noreply.github.com>

* Give ClickHouse a 2m health-check start period

ClickHouse on these runners can take over 50s to start listening
during load spikes (the same slowness behind the Aug 3+ ecto.create
failures). Without a start period the health check flips to unhealthy
after ~50s and the runner kills the job before the server is ready;
probe failures during the start period no longer count toward
--health-retries.

Co-authored-by: Codesmith <codesmith-bot@users.noreply.github.com>

* Probe ClickHouse health check via 127.0.0.1, not localhost

The server fails to bind [::] in these containers (IPv6 disabled) and
falls back to 0.0.0.0 only, while busybox wget resolves localhost to
::1 and does not retry the next address, so the probe got connection
refused forever and every service init failed. Reproduced against
clickhouse/clickhouse-server:25.11.5.8-alpine: localhost fails,
127.0.0.1 succeeds and the container reports healthy within seconds.

Co-authored-by: Codesmith <codesmith-bot@users.noreply.github.com>

---------

Co-authored-by: Codesmith <codesmith-bot@users.noreply.github.com>

Co-authored-by: Codesmith <codesmith-bot@users.noreply.github.com>

…t time) (#6571)

* Revert "Use Blacksmith checkout with sticky-disk git mirror in Elixir CI (#6564)"

This reverts commit fd528f2.

* Fix inert service container health checks in Elixir CI

The postgres and clickhouse health-check options were nested under env,
so Docker never received a health check and jobs could start before
ClickHouse was listening, failing ecto.create with
Mint.TransportError{reason: :closed}. Move options to the service level
and point the ClickHouse check at the real HTTP port (8123 /ping)
instead of the unused 8124.

Co-authored-by: Codesmith <codesmith-bot@users.noreply.github.com>

* Give ClickHouse a 2m health-check start period

ClickHouse on these runners can take over 50s to start listening
during load spikes (the same slowness behind the Aug 3+ ecto.create
failures). Without a start period the health check flips to unhealthy
after ~50s and the runner kills the job before the server is ready;
probe failures during the start period no longer count toward
--health-retries.

Co-authored-by: Codesmith <codesmith-bot@users.noreply.github.com>

* Probe ClickHouse health check via 127.0.0.1, not localhost

The server fails to bind [::] in these containers (IPv6 disabled) and
falls back to 0.0.0.0 only, while busybox wget resolves localhost to
::1 and does not retry the next address, so the probe got connection
refused forever and every service init failed. Reproduced against
clickhouse/clickhouse-server:25.11.5.8-alpine: localhost fails,
127.0.0.1 succeeds and the container reports healthy within seconds.

Co-authored-by: Codesmith <codesmith-bot@users.noreply.github.com>

---------

Co-authored-by: Codesmith <codesmith-bot@users.noreply.github.com>

* make verification work in a banner

* move custom url input into the banner

* improve notice styles

* expandable instructions

* stop polling pageviews

* rely on tracker_script_config object for installation_method

* proper dismiss logic

* [revert me] debugging UI scenarios

* rename verification (component) to verification_banner

* Refine verification banner UI and copy (plausible#6525)

* Refine verification banner UI and copy

- Simplify failure CTAs: keep "Check again" as primary, add ghost
  "Review installation" (or "Try another URL" when a custom URL retry
  is offered); remove installation-guide, change-method, and view-snippet
  expandable logic
- Render inline "verify your installation manually" / "review your
  installation" links inside recommendation text (offer_custom_url_input
  scenarios) via safe HTML helpers, avoiding HEEx whitespace pitfalls
- Tighten diagnostics title and body copy
- Move the "Setup pending" pill in the sites list to replace the
  percentage indicator; use the generic pill component
- Extend the notice component with title_class override and a spinner
  slot; swap success icon to solid check-circle

* improve templating logic

---------

Co-authored-by: Robert Joonas <robertjoonas16@gmail.com>

* emails CTA banner

* revert changes committed for debugging

* fix test after rebase

* fix CE tests

* improve stats_controller_test.exs

* E2E: named verification results + add genserver for mock scenarios

* E2E: checks_mock module

* E2E: plug in the ChecksMock module

* E2E: put_verification_scenario endpoint

* e2e test for verification success

* get CI green

* migration: add onboarding_status

* update site schema with the new field

* advance status to verification_succeeded

* fix setup pending pill condition (/sites page)

* further guard verification banner rendering in provisioning flow

* fix verification re-triggering on refresh (review/domain_change)

* add launch_delay option to MockScenarios

* fix site switcher links + more e2e test coverage

* new logic for email reports CTA

* fix verification banner dismiss button showing through the dashboard options menu

* validate named result exists in MockScenarios.put

* move ChecksMock and MockScenarios into a single file, use :ets

* move checks_mock.ex to test/support

* different success message depending on the flow

* fixup: make sure component goes into loading state instantly after clicking retry

* stats_start_date -> ensure_stats_start_date

...and return the whole site struct. StatsController.stats action needs
the whole site returned to known the "just updated" onboarding status
when rendering the dashboard stats.html.

* advance onboarding status from /sites page too

* push_navigate -> redirect

* changelog + change dashboard link text for CE on installation screen

* credo and remove redundant else clause

* fix installation_test (CE)

* fix E2E test

* remove runtime Mix.env call

* do not persist verification params in browser history

* Add workaround for CH query condition cache bug (plausible#6562)

* Use Blacksmith checkout with sticky-disk git mirror in Elixir CI (plausible#6564)

Co-authored-by: Codesmith <codesmith-bot@users.noreply.github.com>

* Revert to actions/checkout in Elixir CI (git mirror regressed checkout time) (plausible#6571)

* Revert "Use Blacksmith checkout with sticky-disk git mirror in Elixir CI (plausible#6564)"

This reverts commit fd528f2.

* Fix inert service container health checks in Elixir CI

The postgres and clickhouse health-check options were nested under env,
so Docker never received a health check and jobs could start before
ClickHouse was listening, failing ecto.create with
Mint.TransportError{reason: :closed}. Move options to the service level
and point the ClickHouse check at the real HTTP port (8123 /ping)
instead of the unused 8124.

Co-authored-by: Codesmith <codesmith-bot@users.noreply.github.com>

* Give ClickHouse a 2m health-check start period

ClickHouse on these runners can take over 50s to start listening
during load spikes (the same slowness behind the Aug 3+ ecto.create
failures). Without a start period the health check flips to unhealthy
after ~50s and the runner kills the job before the server is ready;
probe failures during the start period no longer count toward
--health-retries.

Co-authored-by: Codesmith <codesmith-bot@users.noreply.github.com>

* Probe ClickHouse health check via 127.0.0.1, not localhost

The server fails to bind [::] in these containers (IPv6 disabled) and
falls back to 0.0.0.0 only, while busybox wget resolves localhost to
::1 and does not retry the next address, so the probe got connection
refused forever and every service init failed. Reproduced against
clickhouse/clickhouse-server:25.11.5.8-alpine: localhost fails,
127.0.0.1 succeeds and the container reports healthy within seconds.

Co-authored-by: Codesmith <codesmith-bot@users.noreply.github.com>

---------

Co-authored-by: Codesmith <codesmith-bot@users.noreply.github.com>

* fight with latency (Try another URL button)

* fight latency (clicking Check again)

Adjust the global btn-base styling to make the button appear "disabled",
when phx-click-loading is set on it.

* nom format e2e

---------

Co-authored-by: Sanne de Vries <65487235+sanne-san@users.noreply.github.com>
Co-authored-by: Artur Pata <artur.pata@gmail.com>
Co-authored-by: Cenk Kücük <cenk@plausible.io>
Co-authored-by: Codesmith <codesmith-bot@users.noreply.github.com>