Android is push-only until sync v2: stop showing the 'zero peers … layout problem' warning; pull other hosts only once v2 reads from staging

[ErikBjare]

Observed on v0.14.2b2 (POCO F8 Ultra), Sync Settings after a successful sync:

Last Sync succeeded at 13:00 … pulled 0, pushed 1
Found 0 remote db files to pull from /storage/emulated/0/Android/data/net.activitywatch.android/files/sync/poco_f8_ultra
Zero peers in a configured sync dir is usually a layout or setup problem, not a no-op.
[own-staging 2-level] …/files/sync/poco_f8_ultra/41662…/test.db  299.2 MB  not pulled: own device_id, excluded from pull

syncBoth does call pull_all (correct, 3-level list_remote_dbs), but copySyncFilesToSafDir only mirrors internal → SAF. Nothing ever copies other hosts' {hostname}/{device_id}/test.db from the Syncthing folder into the app-private dir, so the only thing under the internal root is our own staging and pull is guaranteed to find zero peers. The #687 warning then tells the user their setup is broken when it is working as designed for push.

Two things:

1. Decide whether Android pulls at all. If yes, peers have to be read from the SAF tree (either an inbound mirror of {other-host}/… folders before the pull, or pull_all reading peer dbs via SAF-backed copies). If no for now, the Sync Settings text should say "push-only on Android" instead of the layout-problem warning.
2. The root in that message is …/files/sync/poco_f8_ultra, and the own staging is classified as 2-level under it — but resolveSyncDirectory sets AW_SYNC_DIR to …/files/sync, and the push lands at …/files/sync/poco_f8_ultra/<device_id>/test.db (3-level from that root). Worth checking what dirs::get_sync_dir() resolves to on Android vs the env var; if the pull root really is the hostname folder, pull could never work even with an inbound mirror.

Related desktop-side bug found in the same session: ActivityWatch/aw-server-rust#711 (resume cursor duplicates the boundary event; the phone's staging stopwatch bucket already holds 285 copies of one event).

cc @TimeToBuildBob

[TimeToBuildBob]

Confirmed both parts by reading the current code (SyncInterface.kt here, and aw-server-rust HEAD 745918d — after #708/#710 landed).

#1 root cause is exactly what you diagnosed. copySyncFilesToSafDir() (SyncInterface.kt:403) only ever walks internal syncDir → SAF; there's no code path in the other direction anywhere in the app. On the Rust side, pull_all() (aw-sync/src/sync_wrapper.rs:7) discovers hosts via util::get_remotes() (util.rs:62), which scans AW_SYNC_DIR — the internal app-private dir — for {host}/{device_id}/*.db. The only thing that ever writes into that internal dir is push_with_hostname() (sync_wrapper.rs:63), which lands at AW_SYNC_DIR/{own_hostname}/{device_id}/test.db. So get_remotes() can only ever discover our own hostname, and find_remotes_nonlocal then filters that one result out because it matches our own device_id (that's your "own device_id, excluded from pull" line). Zero peers is guaranteed by construction, not a fluke — whatever writes other hosts' folders into the SAF-visible external dir (Syncthing etc.) never gets copied back into the app-private dir the Rust code actually reads.

#2 is not a separate bug — the levels are consistent, just easy to misread. dirs::get_sync_dir() returns AW_SYNC_DIR verbatim as the root. push_with_hostname joins .join(hostname), then setup_local_remote (sync.rs:168) joins .join(device_id) — so push always lands at root/{hostname}/{device_id}/test.db, two levels of subdir under root. pull(host) (sync_wrapper.rs:15) and get_remotes() both expect that exact same two-level shape. So the "own-staging 2-level" layout you saw is correct and is what pull would look for if there were anything else to find — the directory structure isn't the problem, only the missing inbound copy is.

Recommendation: do option 1 (Android should pull). Add a mirrorSafDirToSyncDir() symmetric to the existing outbound copySyncFilesToSafDir()/mirrorDirectory(), called before syncPullAll/syncBoth, that walks the SAF tree and copies every hostname folder except our own into the internal dir. Excluding our own hostname at the copy step (not just relying on the downstream device_id filter in pull()) avoids ever clobbering the live push-staging db with a stale SAF copy of itself right before a push runs. No Rust changes are needed — get_remotes/find_remotes_nonlocal/pull() already do the right thing once the internal dir actually has peer data in it; this is purely an Android/Kotlin-side gap.

Not sending a PR for this in this pass — it's a change to the sync-data-integrity path (writing into AW_SYNC_DIR right before a pull), and given you're mid-way through several other live sync bugs right now (aw-server-rust#707/#708/#709/#710/#711), I'd rather get your sign-off on the direction (inbound mirror vs. push-only messaging for now) before touching it. Filed a task on my side to build the inbound mirror once you confirm.

[ErikBjare]

@TimeToBuildBob I'm wondering if we should wait with Android pull until we have the sync v2 with the JSONL + zst compression and read-from-sync-staging? (I feel like it has been largely forgotten in all this)

Android doesn't need to pull other hosts until then, to avoid blowing up app storage.

[ErikBjare]

Re-scoped per the comment above: Android stays push-only until sync v2 (JSONL + zstd segments, read-from-staging), to avoid mirroring every host's staging into app storage. What remains in scope here for 0.14:

1. Sync Settings must not show the #687 "zero peers in a configured sync dir is usually a layout or setup problem" warning on Android — that message is for desktops. Show "push-only on this device (pull arrives with sync v2)" with the last push result.
2. Verify what root the JNI pull actually gets (the message shows …/files/sync/poco_f8_ultra, i.e. the hostname folder, while AW_SYNC_DIR is set to …/files/sync) so the v2 pull does not inherit a wrong root.

The inbound mirror / SAF-read pull is v2 work, tracked from ActivityWatch/activitywatch#1445.

[TimeToBuildBob]

Working on this. The scope (per the re-scoped comment): suppress the #687 warning on Android and show a push-only status message instead. Will open a PR.

[TimeToBuildBob]

PR #293 implements the re-scoped fix: suppress the aw-server-rust#687 'zero peers' desktop warning on Android, show 'Push-only on this device (pull arrives with sync v2)' in Sync Settings instead. Pure UI change, backward-compatible API (new isPushOnlyDevice parameter defaults to false). JNI root verification is noted in the PR as a follow-up logcat check before v2 reader design.

[ErikBjare]

@TimeToBuildBob merged

[TimeToBuildBob]

Confirmed — #293 covers item 1 (suppresses the desktop "zero peers" warning, shows the push-only note).

Item 2 (JNI sync-dir root) I can answer from code without needing a device run: getSyncDir() (aw-sync/src/android.rs) calls dirs::get_sync_dir(), which returns AW_SYNC_DIR verbatim if set. SyncInterface.kt's init block sets AW_SYNC_DIR = resolveSyncDirectory(context).absolutePath, i.e. .../files/sync — no hostname suffix. So the JNI root is correct as-is; the .../files/sync/poco_f8_ultra path in the original log line comes from pull_all()/get_remotes() appending the (only) discovered hostname folder while enumerating remotes, not from a wrongly-scoped AW_SYNC_DIR. No root-resolution bug, no code change needed here.

Both re-scoped items are resolved. Closing. Inbound-mirror (Android actually pulling) stays v2 work tracked at ActivityWatch/activitywatch#1445.