yavsc/INSTALL.md
Paul Schneider e72df234bc
Replace MediaMTX with encrypted SRT ingestion via FFmpeg.NET
Manage bounded UDP listeners directly in Yavsc.Api, expose secret SRT URLs in PostIt, and preserve quotas, revocation and replay. Update deployment and privacy documentation, remove relay configuration, and add real-media lifecycle regressions.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
2026-11-08 16:19:41 +00:00

33 KiB

Install

With internet access, Git, dotnet, Node.js, and GNU/Makefile, simply follow these instructions, in order to have the server part installed:

git clone https://github.com/pschneider/yavsc.git
cd yavsc
dotnet restore
npm install          # restore front-end dependencies
npm run build:js     # build the JS/CSS bundles into wwwroot
cd contrib
make install

The front-end bundles (under src/Yavsc.Org/wwwroot/js/*.bundle.min.js and src/Yavsc.Org/wwwroot/css/*.min.css) are produced by npm run build:js and committed to the repo, so the runtime .NET stack does not need a Node toolchain at publish time. Re-run npm run build:js after editing any of the entry files under src/Yavsc.Org/wwwroot/js/ or any of the build/*.entry.js bundle manifests.

About the mobile app,

cd src/PostIt/PostIt.Android
dotnet publish -c Release

and then, install the app, from the src/PostIt/PostIt.Android/bin/Release/net10.0/PostIt.apk

PostIt.Desktop on Windows (per-user installer)

Build prerequisites: the .NET 10 SDK and NSIS 3 (with its standard Modern UI 2 includes). From the repository root, publish a self-contained Windows x64 application, then package the entire publish directory:

dotnet publish src/PostIt/PostIt.Desktop/PostIt.Desktop.csproj -c Release -r win-x64 --self-contained true -p:PublishSingleFile=false
makensis /DAPP_VERSION=1.1.0-estimate.1 contrib/Windows/postit.nsi

Set APP_VERSION to the version of the published application (the Version property in PostIt.Desktop.csproj). NSIS requires this value; it is used in the installer filename and Windows' installed-apps list. On Linux, use makensis -DAPP_VERSION=1.1.0-estimate.1 contrib/Windows/postit.nsi instead. Cross-publishing uses the same dotnet publish command. If makensis is not on PATH on Windows, invoke the executable from the NSIS installation directory.

The result is contrib/Windows/PostIt-<version>-win-x64-Setup.exe. NSIS also accepts PUBLISH_DIR (an absolute path to a different win-x64 publish directory) and OUTPUT_FILE (the output executable path), with the same /DNAME=value or -DNAME=value syntax. Quote an entire define argument when its value contains spaces.

The installer does not request elevation or install a shared .NET runtime. It installs in %LOCALAPPDATA%\Yavsc\PostIt.Desktop, creates a Start menu shortcut for the current user and registers postit:// in HKCU\Software\Classes so browser-based sign-in returns to PostIt. Windows' installed-apps list provides the uninstaller. Close PostIt before installing an update or uninstalling it.

The app subdirectory is reserved for published application files and is removed recursively on uninstall. User settings and tokens stored outside this directory are preserved. Reinstalling uses the recorded installation directory and overwrites the packaged files.

The self-contained publication includes .NET, but does not add native LibVLC libraries: video playback still requires a compatible Windows x64 LibVLC installation. PostIt reports an unavailable video player and offers the system browser as an alternative.

Before distributing a release, test installation, startup, the postit:// login callback, updating and uninstallation on Windows with a standard (non-administrator) account. Installer compilation alone does not validate these Windows runtime behaviours.

Resource consumption and CGU

Account withdrawal and payment retention

Apply the AccountWithdrawal migration before deploying all three hosts. It adds withdrawal timestamps and replaces the immutable alternate email key with a required unique email index, preserving uniqueness while allowing pseudonymization. Do not run the withdrawal flow until every host uses the matching backend.

Yavsc.Org must have write/delete access to the same managed volumes as API/Blogs: Site:UserFilesRoot, Site:Avatars and Podcast:RootPath. Use absolute paths shared across processes; defaults relative to application directories are not shared automatically. Accounting files under Site:Bills are not erased by withdrawal. Symbolic links in managed user directories are refused rather than followed.

The account is blocked before cancellation/erasure, and withdrawal is only confirmed when these operations and pseudonymization succeed. On an explicit failure, an administrator must fix the cause and retry the existing administrative withdrawal action. Ensure no live upload/encoding operation is still writing to the user's volumes during withdrawal; drain such operations before retrying.

The Org hosted worker purges eligible technical payment records at startup and checks at most every minute, with an earlier wake-up for a known anniversary. Monitor errors and service availability: scheduled processing cannot ensure an exact wall-clock deletion while the host is stopped. At least one Org instance must run, and an external operations process must enforce the one-year retention deadline during prolonged downtime. PostIt administrators can also confirm and run POST /api/v1/administration/payment-residues/purge. The report counts accounts, subscriptions, monthly payments, donations and legacy PayPal payments. Active accounts and unfinished withdrawals are never purged. No new subscription credits may be granted after withdrawal begins.

Invoice payments, invoices and other legal accounting records are excluded. Define and publish the actual accounting, logs and backup retention policy for the deployment, verify configured providers and international-transfer safeguards, and prevent withdrawn data from being reactivated by backup restoration.

Configure the same ResourceUsage section on Yavsc.Api, Yavsc.Blogs and Yavsc.Org. API and Blogs must also use the same application database. All three hosts and configuration templates inherit the backend defaults from ResourceUsageOptions. Deploy matching backend versions across hosts. The public CGU renders the effective coefficients and limit. The default limit is 1,000 points. The following is an example explicit override, not a duplicate of defaults. Use the account's immutable user ID (sub), not its login, for overrides:

{
  "ResourceUsage": {
    "ApiCallWeight": 0.2,
    "CpuSecondWeight": 0.1,
    "BandwidthMbWeight": 0.05,
    "StorageMbWeight": 0.5,
    "DefaultLimit": 1000,
    "UserLimits": {
      "example-account-id": 20000
    }
  }
}

Environment variables follow ASP.NET's convention, for example ResourceUsage__DefaultLimit and ResourceUsage__UserLimits__example-account-id. Restart all three services after changing the rule. Invalid weights or limits fail startup. Existing cumulative ResourceUsageRecords are reused, and changing weights recalculates their gross indicator. Apply the ResourceSubscriptions migration before deploying any of the three hosts: it adds the credit balance and subscription/payment tables without clearing existing counters.

Consumption is in points, independent of EstimatedCompensation and currency. Counters have no automatic reset. Storage means cumulative instrumented writes less individually tracked bytes removed or replaced, not total filesystem occupancy. Replacements refund their tracked previous size and charge the new size. Legacy files without an individual receipt receive no refund; storage counters never become negative. Deletion remains available at the consumption limit, but API/transfer/CPU costs are not refunded. Apply the RefundableResourceStorage migration before deployment; its empty receipt table deliberately does not infer provenance from historical counters. Renames preserve receipts and repeated PDF generation replaces the previous tracked volume rather than charging the whole file again. The historical CPU counter is instrumented elapsed processing time, not exhaustive CPU profiling.

Authenticated HTTP operations are admitted atomically against known costs (API call and declared request body). Personal-file uploads additionally check their aggregate storage cost before writing. Response bytes and unknown-length request bodies are measured through counting streams, and remaining costs are recorded after execution, including failed requests. PostgreSQL transaction-scoped advisory locks serialize admission and accounting for each user across hosts. In-flight operations can exceed the limit through costs unknown at admission; subsequent operations return HTTP 429 with resource_quota_exceeded Problem Details. No permanent account suspension or strict running-operation interruption is performed.

GET /api/v1/resource-usage/me returns Consumption, ConsumptionLimit, RemainingConsumption and IsQuotaReached, alongside existing counters. It also returns GrossConsumption, AppliedCredits and CreditBalance. Consumption is the gross indicator less acquired credits, floored at zero. RemainingConsumption includes both unused free quota and carried credit. Usage consultation (including the role-protected admin overview), login, session refresh and the CGU remain accessible without consuming quota. The quota does not measure local PostIt resources, direct peer calls or WebSocket traffic. HTTP upload size limits and disk quotas still apply.

The public page /Home/CGU describes the effective rule and default limit (and the individual limit for a signed-in user). It is linked from the Web footer and PostIt's About page, using the configured authentication authority. PostIt displays consumption and remaining points in My profile and the administration usage views.

Monthly resource subscriptions (PayPal REST)

The free registration keeps the configured cumulative quota without a monthly reset. Discovery costs EUR 1 excluding VAT (EUR 1.20 including 20% VAT) per month; Premium costs EUR 2 excluding VAT (EUR 2.40 including VAT). Every confirmed payment grants up to one / three default quotas, respectively, with renewal on the UTC monthly anniversary of the PayPal start date. The pack allowance is captured at subscription creation, not multiplied by an individual user's quota override.

At reconciliation, old carried credit is used first; the new allowance erases remaining net usage. Two thirds of the unused allowance are added to permanent carried credit, rounded down to eight decimal places. Old carried credit never loses another third and survives cancellation. Example: 400 points of usage and 1,000 new points erase 400 and carry 400; an idle Premium month carries 2,000 points. Counters are never deleted. Pending/failed payments give no credit. Refunds (including partial refunds) revoke the corresponding award once.

Discovery grants twice Constants.DefaultFSQ (1,000 MiB), and Premium five times Constants.DefaultFSQ (2,500 MiB) of disk space during each confirmed, non-reversed paid monthly period. Free registrations use the standard quota. The reference is the standard free quota (500 MiB), not an individual's administratively adjusted DiskQuota. That base quota is never overwritten. New user objects now use the existing 500 MiB database default rather than the former 512 MiB CLR initializer; existing accounts retain their stored quota. Cancellation retains the disk allowance through the UTC anniversary ending the paid period; expiry or refund restores the base quota without deleting existing files. Uploads exceeding available space are refused before writing. The maximum size of one file and separate podcast limits are unchanged. Pending/failed payments grant no disk bonus. Apply the ResourceSubscriptions, PremiumDiskQuota and DiscoveryDiskQuota EF migrations to the shared database before deploying any host. Profile and usage responses expose the effective disk quota; usage also exposes PremiumDiskQuotaUntil and DiscoveryDiskQuotaUntil. When paid periods overlap, Premium takes precedence until its expiry, then any still-paid Discovery period applies.

Configure only Yavsc.Org for PayPal subscriptions. API and Blogs need the migrated shared database and matching resource weights, not PayPal credentials. Existing Classic donation settings remain unchanged. Use the existing PayPal:Mode, PayPal:ClientId and PayPal:ClientSecret REST settings; keep credentials in user-secrets or environment variables.

{
  "ResourceSubscriptions": {
    "DiscoveryPlanId": "P-REPLACEWITHDISCOVERYPLAN",
    "PremiumPlanId": "P-REPLACEWITHPREMIUMPLAN",
    "WebhookId": "REPLACEWITHWEBHOOKID",
    "DiscoveryPriceExcludingTax": 1,
    "PremiumPriceExcludingTax": 2,
    "VatPercent": 20
  }
}

Before enabling checkout, create a PayPal subscription product and two ACTIVE billing plans in the same REST app and environment: one indefinite REGULAR cycle with interval_unit: MONTH, interval_count: 1, total_cycles: 0, and EUR fixed_price.value of 1.20 / 2.40 respectively. These are already VAT-inclusive amounts: do not add PayPal tax, setup fees, trials, quantity support or shipping fees. Disable billing outstanding balances in the plans (auto_bill_outstanding: false) to avoid bundling missed installments into an amount that does not match one monthly payment. Checkout checks the configured plan against its expected TTC amount and refuses automatic collection of outstanding balances. Numerically invalid pricing fails startup; missing plan IDs/webhook/ credentials returns an explicit unavailable error instead of a fake success.

Register the HTTPS endpoint https://<Yavsc.Org host>/ResourceSubscription/Notify in that REST app. Subscribe to BILLING.SUBSCRIPTION.ACTIVATED, UPDATED, CANCELLED, SUSPENDED, EXPIRED, PAYMENT.FAILED, and PAYMENT.SALE.COMPLETED, REFUNDED, REVERSED (the lifecycle events use the BILLING.SUBSCRIPTION. prefix). Set the returned webhook ID in configuration. The server verifies the signature through PayPal, then fetches the bound subscription and transaction history; it never grants credit from an unverified webhook or the browser return. Payment IDs and monthly cycles are unique, and the same per-user PostgreSQL lock serializes usage and awards. Delayed notifications take effect on the usage measured when reconciled.

Users manage subscriptions at /ResourceSubscription, linked from the Web footer and PostIt's About/profile pages. A browser Web login may be required independently of PostIt's session. Create/refresh/cancel require antiforgery tokens and ownership; the management page remains available at the quota. Refresh payments reconciles missed notifications and historical refunds. Cancellation is confirmed with PayPal, not merely recorded locally. Only one non-terminal subscription is allowed per account. An uncertain creation is retried with its same PayPal request ID for at most 24 hours; an older unresolved creation requires merchant-side review before allowing a second checkout. Do not delete that record and retry without checking PayPal for a possible existing agreement.

Run a real Sandbox approval, first payment, renewal, repeated webhook, refund and cancellation before enabling production. The repository tests use synthetic PayPal transports and do not validate merchant credentials or PayPal account eligibility.

Podcast live broadcasts and replays

OBS Studio for planned emissions

OBS publishes encrypted SRT directly to Yavsc.Api, using FFmpeg.NET (xFFmpeg.NET) to manage FFmpeg listeners. No MediaMTX, external media server, authentication callback or RTMPS certificate is required. Install FFmpeg with libsrt, H.264/AAC and MPEG-TS support on the single Yavsc.Api media host (ffmpeg -h protocol=srt). OBS supports SRT in its Custom streaming service; see its SRT streaming guide. This replaces the previous RTMP/RTMPS transport, not the authenticated playback API.

Apply the PodcastObsIngestion migration on the shared database. Configure Podcast:Obs on Yavsc.Api only, for example:

{
  "Podcast": {
    "Obs": {
      "Enabled": true,
      "ServerUrl": "srt://obs.example.org",
      "ListenAddress": "0.0.0.0",
      "FirstPort": 10000,
      "PortCount": 16,
      "ActivationMinutes": 30,
      "MaxDurationMinutes": 120,
      "MaxConcurrentBroadcasts": 2
    }
  }
}

ServerUrl contains the public SRT host only, with no port, path, secret or query. ListenAddress is a literal local IP address (IPv4 or IPv6); configure a matching public address family. Open/forward UDP 10000-10015 in this example. An ordinary HTTP reverse proxy does not carry SRT. NAT forwarding must preserve the advertised port numbers. One FFmpeg process and UDP port are reserved per armed emission, even before OBS connects. Armed and running emissions share MaxConcurrentBroadcasts; an exhausted pool returns HTTP 503. The port range is bounded (at most 128 ports) and must be large enough for the concurrency limit. Listener startup failure is reported, not treated as successful activation. Keep process arguments and credentials accessible only to the service administrator. Limit ingress bitrate at the network level and in OBS: application quotas are checked on completed media segments, not on every incoming packet.

In PostIt:

  1. Prepare an emission and select it.
  2. Choose Activer OBS / renouveler la cle.
  3. In OBS, select Parametres > Stream > Service : Personnalise, then copy the complete displayed secret SRT URL as Server, leaving Stream Key empty. The URL includes caller mode, a random passphrase and enforced AES-256 encryption. Use H.264 video, AAC audio, a 2-second keyframe interval and a bitrate producing segments below 8 MiB.
  4. Start streaming in OBS before the activation expires. The actual emission starts when the first complete media segment arrives, even if the scheduled date is later. No camera capture in PostIt is required; PostIt may be closed afterwards.
  5. Stop OBS or choose Revoquer OBS et arreter le diffuseur in PostIt. Committed segments remain available and the normal MP4 worker prepares replay.

Passphrases are scoped to one emission, displayed once and rotated on activation. Only a SHA-256 hash and lifecycle metadata are stored in the database; FFmpeg needs the actual passphrase in memory/process arguments for SRT encryption. It is never included in status responses or persisted in PostIt. Navigation/logout clears the displayed secret URL but does not revoke a running external emission. Revocation is available at the consumption limit. After disconnect, restart or timeout, the emission ends rather than silently resuming under a new publisher: prepare a new emission for another broadcast. An API restart invalidates all old activations; an unstarted emission must be activated again because its passphrase is not recoverable from the database. An invalid handshake can end an armed listener; no podcast is created without media, and the owner can activate again. Rate-limit connection attempts at deployment.

The worker manages bounded FFmpeg.NET listener/remuxing tasks. It rejects ended, deleted, revoked or withdrawn accounts and caps concurrent captures and duration (the lower of planned duration and configured maximum). Revocation cancels the corresponding FFmpeg process and frees its port. Activation expiry stops unused listeners. The passphrase is required before SRT media can be received; a stream name alone is not an authentication mechanism. Committed segments are charged as transferred and stored MiB under the same consumption rule; their storage receipts are refunded on podcast deletion. OBS storage counts towards the user's effective disk quota and the existing podcast media limits. Pending capture files are monitored and bounded; enforcement can overshoot by one polling interval of incoming data. Processing errors are reported in the owner-only OBS status endpoint, without exposing secret URLs or FFmpeg diagnostics. Monitor process/cleanup failures and drain active captures before maintenance. An API outage ends ingestion and requires reactivation.

The normal HTTP API authenticates playback; OBS stream credentials never grant viewer or account access.

For the real media regressions, install FFmpeg with libsrt and run:

dotnet test --project src/test/Yavsc.Api.Tests/Yavsc.Api.Tests.csproj --filter-class '*PodcastObsTests'

The tests launch loopback-only listeners, generate synthetic H.264/AAC media with FFmpeg, verify encrypted SRT through FFmpeg.NET, wrong-passphrase rejection, revocation during streaming, AAC audio, HLS segments, MP4 replay and refund, then stops its own processes and removes its temporary files.

PostIt Desktop's home page provides Podcasts : directs et replays. Install LibVLC 3 with the livehttp, MPEG-TS, x264 and AAC plugins on the Desktop machine. Windows capture uses DirectShow; Linux uses V4L2 and the default PulseAudio microphone. FFmpeg is not a client dependency. Select a camera and optionally a microphone, enter a title, then start. An Apercu local de la capture appears at the top of the Podcasts page on Android and Desktop. It uses the current capture, not a second camera session or the server replay. Android displays the same oriented frames as the encoder, with black bars to preserve proportions; Desktop duplicates LibVLC's video output. Local microphone playback is disabled to avoid acoustic feedback. Stopping releases the preview; hiding/recreating the Android preview surface does not restart the encoder or change its timestamps. Stopping publishes the retained replay; navigating away or logging out stops capture too. A second signed-in PostIt instance can watch the live broadcast while it is being captured.

Both upload and playback require a bearer token with the api scope. All authenticated API users can watch the catalogue, live broadcasts and replays; only the broadcaster can upload, complete or delete their podcast. Media is not anonymous and no access token is included in a playback URL.

Configure Yavsc.Api's Podcast section (or environment variables such as Podcast__RootPath). The base media store needs no database migration; planned emissions and OBS require their migrations documented above. Install FFmpeg 5+ on the API server (with concatf, MPEG-TS, MP4 and AAC bitstream-filter support; for Debian/Ubuntu: apt-get install ffmpeg). The API runtime stage in the main Dockerfile includes it. Only stream-copy remuxing is performed, not video/audio re-encoding:

{
  "Podcast": {
    "RootPath": "/srv/yavsc-private/podcasts",
    "MaxSegmentBytes": 8388608,
    "MaxOwnerBytes": 1073741824,
    "MaxSegments": 7200,
    "MaxOwnerPodcasts": 100,
    "IdleTimeoutSeconds": 60,
    "FfmpegPath": "ffmpeg",
    "ReplayTimeoutSeconds": 600
  }
}

RootPath must be absolute, writable by Yavsc.Api and outside any static file directory. The default is DataDir/podcasts under the API application base directory. Use a persistent volume and back up this entire directory (indexes and segments together). This implementation supports one API process per storage root, not multiple writers or replicas.

Defaults allow 8 MiB per segment (also the HTTP endpoint's hard limit), 1 GiB retained per owner, 7,200 segments per episode and 100 episodes per owner. At the target two-second segment duration, this is roughly four hours per episode; the byte quota can be reached earlier. Podcast quotas are separate from the legacy user-files DiskQuota for native HTTP capture. OBS source segments also count towards the user's effective disk quota and instrumented consumption; the additional MP4 is counted only by MaxOwnerBytes. Deleting an ended podcast releases its podcast quota. Reverse proxies must allow an 8 MiB PUT body and forward Authorization, Range and Content-Range on every API request/response. The retained MPEG-TS and the additional MP4 count toward MaxOwnerBytes. Allow enough free quota and disk space for both copies and the temporary MP4; deletion removes all copies.

A broadcast without a new committed segment for 60 seconds, or present when the API restarts, becomes an interrupted replay on the next catalogue, manifest or upload request. Already committed media remains playable. The broadcaster retries transient upload errors twice using the same sequence and content; it stops if capture fails or the pending local media exceeds 64 MiB. Unacknowledged media is retained under %LOCALAPPDATA%\PostIt\podcast-capture (Linux: the corresponding local application data directory), with podcast.json and the capture playlist for manual recovery. There is no automatic resumption after an application restart; these private local recordings may be removed manually when no longer needed.

On completion, the API queues a private, seekable MP4 (-c copy, +faststart). The persisted replay.state moves through Pending, Processing, Ready or Failed; completion itself does not wait for FFmpeg. The single API worker also queues older Completed indexes automatically, recovers interrupted preparation after a restart, and scans every five seconds. Interrupted broadcasts keep MPEG-TS playback unless the owner explicitly completes them. Source segments and existing HLS routes remain available. Preparation failure keeps all source media and reports an error; after checking server logs, FFmpeg installation, timeout and quota, the owner can use PostIt's Relancer la preparation MP4 action, or POST /api/v1/podcasts/{id}/replay/retry. The MP4 is exposed only when ready, at authenticated GET /api/v1/podcasts/{id}/replay.mp4 (also HEAD; byte ranges supported).

PostIt's Android and Desktop player waits for MP4 preparation, then fetches bounded authenticated ranges with token refresh rather than giving LibVLC a bearer URL. Replay controls overlay the native video: play/pause, position, duration and draggable seek bar. They hide after three seconds without interaction while playing, stay visible during seeking/pause/keyboard focus, and reappear on pointer movement or touch. The live player still joins up to three segments behind the edge. Native camera/microphone permissions, drivers and codecs must still be verified on each target Windows/Linux machine.

Android podcasts

The Android APK bundles LibVLC and uses a native Android video surface inside the same Avalonia video page as desktop. Catalogue, playback controls and authenticated segment downloads are shared; no external browser, public media URL or camera/microphone permissions are needed to watch a podcast.

Video playback keeps the screen awake while PostIt is visible. Pausing, stopping, reaching the end, a playback error or leaving the video releases this request; backgrounding the app always allows the screen to sleep.

The podcast page offers Garder l'ecran allume pendant l'apercu camera et le direct on Android. This preference is saved and disabled by default. It only keeps the screen awake during an active camera preview or broadcast while PostIt is visible, and can be changed during capture. Turning it off does not stop capture or alter the existing background broadcast service.

Leaving the app stops playback and cancels downloads. Returning recreates the player: a replay starts from the beginning, and a live stream rejoins near the live edge. Seeking and background audio are not supported.

Broadcasting requires Android 8 (API 26) or newer. In the shared podcast page, detect devices, select front/rear camera and optionally the microphone, then start the live broadcast while PostIt is visible. Camera permission, microphone permission when selected, and notification permission on Android 13+ must be granted. A permission denial is reported in the page without starting capture or creating a server broadcast. Permissions are requested before the server session is created so a long permission dialog cannot expire it. Watching does not request these permissions.

Android uses Camera2 and its native MediaRecorder H.264/AAC encoder, up to 480p at 30 fps, not the desktop LibVLC camera inputs or software encoders. MPEG-TS is segmented at IDR frames, with timestamps used for segment durations. Closed segments enter the same ordered upload/retry/quota/replay pipeline as desktop. Devices without a supported camera/encoder report a startup error.

Camera frames are rotated through an OpenGL surface before encoding, so both live playback and replays have upright pixels without relying on rotation metadata (not supported by the MPEG-TS recorder). The sensor, front/rear lens and display orientation are sampled when capture starts; portrait output swaps width and height. Camera2 may already include sensor rotation in the SurfaceTexture transform; only the remaining rotation is applied by OpenGL. Keep that device orientation during a broadcast, or stop and start a new broadcast after turning the device.

A camera/microphone foreground service and a partial wake lock maintain the broadcast when PostIt is backgrounded or the screen locks. Its ongoing notification returns to the app and offers Arreter et publier. Stopping flushes the final segment before completing the replay. Leaving the podcast page or logging out still stops capture. Android's explicit service stop, process termination or application restart does not automatically restart a camera; segments already received by the server remain available under the interrupted-replay rules above. Device power policies can impose additional restrictions and must be tested on the intended physical devices.

Native library compatibility: the Android app now uses VideoLAN.LibVLC.Android 3.7.0-beta, which is the minimum compatible package for 16 KiB-page devices. Older 3.6.5 binaries can fail at runtime with failed to load libc++_shared.so or the linker error XA0141 because the native libraries are not aligned for 16 KiB memory pages. A successful APK build or a test on a 4 KiB-page emulator does not establish compatibility for 16 KiB-page devices or Play distribution requiring that alignment.

Build with the Android workload and SDK configured:

dotnet build src/PostIt/PostIt.Android/PostIt.Android.csproj

On an Android device, verify replay video/audio, live playback, pause/resume, rotation, Back navigation and returning after backgrounding the app. Also verify an expired session or interrupted network reports a playback error. For broadcasting, test front/rear cameras, microphone enabled/disabled, permission denial, screen lock, background capture, notification stop and the final replay. No automatic retry of a denied permission or a killed camera service is performed.

The Debug APK includes a local native-capture test activity (excluded from Release); it does not create a server podcast. On a dedicated emulator, grant camera, audio and notification permissions, then run:

adb shell am start -W -f 0x18000000 -n fr.pschneider.postit/fr.pschneider.postit.PodcastCaptureTestActivity
# Wait about 20 seconds for foreground/background recording and final flush.
adb shell run-as fr.pschneider.postit cat files/podcast-capture-test-result.txt
# Repeat with the second camera, no audio, and the notification stop action:
adb shell am start -W -f 0x18000000 -n fr.pschneider.postit/fr.pschneider.postit.PodcastCaptureTestActivity --es microphone false --es camera front --es notificationStop true

The result reports the segment count, the background progress and the private recording directory under files/podcast-capture-tests. Recordings remain there for local decoder inspection and can be removed after the test.

The implementation was checked on an Android 16 x64 emulator with 4 KiB pages: camera/microphone capture produced five independently decodable H.264/AAC segments across foreground/background operation; the second camera without audio produced six decodable H.264 segments, including screen-off operation and stop through the notification action. The foreground service was gone after both tests. These local tests do not replace an authenticated Android-to-server broadcast/replay test or validation on physical devices.

Resources

System resources needed to install.

Some TCP/IP ports to listen

The following provided services need all a port to listen :

  • Org
  • Blogs
  • Api (to become obsolete, in favor of Performer + Client + Com + Moderation )

A Software bloc name

In this doc, let's say, Yavsc

A domain name

Wait ... pschneider.fr You'll have to see with your provider, in order to power it, and make it point to your host ip, at least concerning the ports 80 & 443.

A Postgresql db

The database must be created, and we need its connection string, allowed to modify the data dictionary, see the .env file

An Smtp mail service

You'll need to send e-mail ... using smtp and the provider of your choice, see the .env file.

Google

I'll have to make with it, a least a moment, it needs a service account, and some setup, in the appsettings.*.json file, and the execution environment variable GOOGLE_APPLICATION_CREDENTIALS to a value pointing the json description file for the Google Service account.