roadmap
- 01
- 02
- 03
- 04
- 05
- 06
- 07
- 08
Prepare for Media3 interviews on how ExoPlayer streams audio and video on Android and keeps playing in the background. Covers HLS and DASH, track selection, MediaSession, caching and downloads.
17 of 17 questions in your plan: 0 known, 0 in progress, 0 due, 0 skipped, 17 new. 0 cut, outside your plan.
Media3 rarely gets a round of its own. It surfaces inside the Android technical interview at companies whose product plays something, from a short-video feed to podcasts, and in a mobile system design round that asks for a video feed and then wants the design mapped onto a real player. The opening questions are about player state: what `prepare()` starts that building the player did not, why `playWhenReady` can be true while nothing moves (the player is still in `STATE_BUFFERING`), and why every call must come from the one thread the player is bound to. Levels separate on where an answer stops. Building an `ExoPlayer`, handing it a `MediaItem` and attaching a `PlayerView` is the entry bar. Past it, candidates explain why audio stops when the user leaves the app unless playback lives in a `MediaSessionService`, a foreground service of type `mediaPlayback` that the UI reaches through a `MediaController`; why a download cache uses `NoOpCacheEvictor`; and why a feed that gives every row its own player runs into the device's limit on hardware decoders. From senior level the questions arrive as bug reports: a live stream paused too long fails on resume, and the answer names `ERROR_CODE_BEHIND_LIVE_WINDOW` and the fix, a seek back to the live edge and a fresh `prepare()`.
The plan opens away from Android on purpose. [Adaptive streaming](/topics/found-system-design-cdn-edge-media-streaming) explains what a player actually fetches: an HLS or DASH manifest, short segments and a ladder of bitrates. Without it, track selection reads as a list of setters; with it, capping resolution on a metered connection becomes a choice about which rungs the player may use. The second stretch is the client design those streams feed, framed as a design round frames it: a [scrolling video feed](/topics/mob-system-design-media-pipelines-video-playback) that preloads the next items, autoplays muted only when visible, reuses a small pool of players and keeps downloads playable until their licence expires. It is platform-neutral, and it comes before the API so that each part of the library arrives with a job waiting for it. [Media3 itself](/topics/mob-media3) closes the plan, and much of it is that design turned into classes: preload depth into `DefaultPreloadManager`, the disk cache into `SimpleCache` behind `CacheDataSource`, offline viewing into `DownloadManager` and its service. The one piece the feed never asked for is the session, which lets the lock screen, a headset or a car control playback, so background audio is learned here rather than in the design.
An Android developer who has never shipped playback should begin at the last module and use the first two as its reference: build one screen that streams video and keeps audio going in the background, then read upward to learn what each setting you touched decides. Engineers from video infrastructure have the opposite gap. Manifests and bitrate ladders are home ground, and the hours belong to the session, the service and the activity lifecycle. If the interview is days away, the one control worth moving is level, narrowed to the band on the posting, since the entry questions cost little to anyone who has built a player. The plan stops at playback. Editing with Media3's Transformer, casting and ad insertion are outside it, and so is the platform beneath the player: activities, Compose and the background-work APIs. The [Android Developer roadmap](/roadmaps/android) carries that platform and the architecture questions asked around a media feature. On iOS the same job falls to AVFoundation, on the [iOS Developer roadmap](/roadmaps/ios).
Media3 rarely gets a round of its own. It surfaces inside the Android technical interview at companies whose product plays something, from a short-video feed to podcasts, and in a mobile system design round that asks for a video feed and then wants the design mapped onto a real player. The opening questions are about player state: what prepare() starts that building the player did not, why playWhenReady can be true while nothing moves (the player is still in STATE_BUFFERING), and why every call must come from the one thread the player is bound to.
Levels separate on where an answer stops. Building an ExoPlayer, handing it a MediaItem and attaching a PlayerView is the entry bar. Past it, candidates explain why audio stops when the user leaves the app unless playback lives in a MediaSessionService, a foreground service of type mediaPlayback that the UI reaches through a MediaController; why a download cache uses NoOpCacheEvictor; and why a feed that gives every row its own player runs into the device's limit on hardware decoders. From senior level the questions arrive as bug reports: a live stream paused too long fails on resume, and the answer names ERROR_CODE_BEHIND_LIVE_WINDOW and the fix, a seek back to the live edge and a fresh prepare().
The plan opens away from Android on purpose. Adaptive streaming explains what a player actually fetches: an HLS or DASH manifest, short segments and a ladder of bitrates. Without it, track selection reads as a list of setters; with it, capping resolution on a metered connection becomes a choice about which rungs the player may use.
The second stretch is the client design those streams feed, framed as a design round frames it: a scrolling video feed that preloads the next items, autoplays muted only when visible, reuses a small pool of players and keeps downloads playable until their licence expires. It is platform-neutral, and it comes before the API so that each part of the library arrives with a job waiting for it.
Media3 itself closes the plan, and much of it is that design turned into classes: preload depth into DefaultPreloadManager, the disk cache into SimpleCache behind CacheDataSource, offline viewing into DownloadManager and its service. The one piece the feed never asked for is the session, which lets the lock screen, a headset or a car control playback, so background audio is learned here rather than in the design.
An Android developer who has never shipped playback should begin at the last module and use the first two as its reference: build one screen that streams video and keeps audio going in the background, then read upward to learn what each setting you touched decides. Engineers from video infrastructure have the opposite gap. Manifests and bitrate ladders are home ground, and the hours belong to the session, the service and the activity lifecycle. If the interview is days away, the one control worth moving is level, narrowed to the band on the posting, since the entry questions cost little to anyone who has built a player.
The plan stops at playback. Editing with Media3's Transformer, casting and ad insertion are outside it, and so is the platform beneath the player: activities, Compose and the background-work APIs. The Android Developer roadmap carries that platform and the architecture questions asked around a media feature. On iOS the same job falls to AVFoundation, on the iOS Developer roadmap.