District published Endgame showtimes, the monitor alerted, and then the shows were
pulled. Dedupe remembered every session id it had announced, and re-listings often
reuse the same ids, so the moment bookings actually opened would have been silent.
It only re-armed this time because a redeploy happened to wipe the state file.
A session missing for three consecutive successful checks is now forgotten, so its
return alerts again. Three rather than one, so a flaky response can't produce a
duplicate max-priority push; failed checks don't count. When the whole date goes
dark after opening, a separate notice says the shows were withdrawn. The re-arm
lives in the base state rather than that notice's patch, so a lost notification
can't leave the monitor disarmed.
A leftover v1 MOVIE_CODE used to switch the whole District block to the old names,
silently discarding DISTRICT_CONTENT_ID even when it was set deliberately. The
monitor then watched the wrong film and still reported 'closed', which is the
failure mode this project exists to avoid.
Each field now prefers the DISTRICT_* value and falls back to its v1 name, an
explicitly emptied value is respected, and any v1 name still in use is warned
about at startup. District is enabled by DISTRICT_CONTENT_ID rather than the
movie code, which is optional: District answers without it, and a film with no
showtimes yet has none.
Adds a 'watching' block with the movie id, variant, venue and format filter per
provider. A wrong id or an unexpectedly narrow filter was previously impossible
to tell apart from 'nothing on sale yet', since both just report closed.
A film that is announced but has no showtimes yet answers 400 with code 32, which
is exactly the state you are monitoring for before a release. Counting that as a
check failure meant pointing District at an upcoming film produced a false BROKEN
alert every six minutes until the listing appeared.
Code 32 now reports closed and carries a note through to the status endpoint and
the daily heartbeat, so a typo'd content id still surfaces rather than looking
like a quiet 'not open yet' forever. Other 400s, including the missing-parameter
error, still fail loudly.
Additive: the old flat District variable names still work, so v1 configs run
unchanged. Two things to know when upgrading are called out in the release notes.
BookMyShow blocks datacenter IPs on top of TLS fingerprinting, confirmed by CI
runners getting 403 with a working impersonation, so cloud hosts need the proxy
for BMS as well as District. impit resolves DNS locally for socks5:// and hands
an IPv6-only exit an unreachable IPv4 address, so the scheme is normalized to
socks5h://. Verified end to end: a proxied BMS check returns 200 and alerts.
Docs said the proxy did nothing for BookMyShow, which was wrong.
BookMyShow sits behind Cloudflare bot management that rejects curl and Node fetch
with 403 on the TLS fingerprint alone, so no header or cookie copying helps and a
proxy makes no difference. impit performs the request with a real Chrome
fingerprint; it ships musl and arm64 binaries, so the Alpine multi-arch image is
unaffected. BMS has no 204 either: a date that is not on sale returns 200 with no
showtimeWidgets key, so 'closed' is read from the payload.
Sites now live in src/providers/, each exposing check() and extract() and returning
one normalized session shape. State is namespaced per site, so each keeps its own
dedupe and failure tracking and neither can silence the other. Alert titles name the
site. Sessions without exact seat counts render without them, which BMS needs.
Old flat District variable names still work. CI now starts the built image and runs
a real check through both providers before publishing, since a broken native binary
only fails at runtime.
GitHub Actions builds linux/amd64 and linux/arm64 on every push to main and
pushes a single :latest tag to ghcr.io, so the container can be run without
cloning or building. Tests gate the publish step.
Aculix Technologies