sync: put the revocation's timeout on the request itself

withTimeoutOrNull around the revoke bounded nothing. The call parks on a
socket read that neither coroutine cancellation nor Thread.interrupt can
break, and withContext returns only when its block does -- so the
deadline passed and we waited anyway, for the shared client's own
budget: 30s per resolved address, doubled by the authenticator's retry,
plus up to 120s of read timeout. Removing a homelab account off the VPN
sat there for minutes with nothing visibly happening. Only closing the
socket ends it, which is what callTimeout does.

The budget moves to AppPassword.revoke, where the "best effort, must not
block the removal" contract is already written down, and where it can be
enforced. Not on the shared client: callTimeout covers the whole
exchange including the body, and a multiget of a large list over a slow
link legitimately runs long. CalDavHttpTest pins that decision.

The destructive tail is now uncancellable. It spans four stores that
cannot share a transaction, and the caller is a viewModelScope tied to
the Settings destination, so a couple of back gestures used to kill it
mid-sequence. Only the DataStore writes can observe cancellation -- every
Room DAO here is blocking -- so the landing point was cadence.forget:
the app password already revoked while the row survives holding it, and
the account asking the user to sign in again for a credential we
invalidated ourselves. Further in, the tasks are gone and the row stays.
The tail is bounded and sub-second, so finishing it always beats
stopping inside it.
This commit is contained in:
2026-09-07 22:37:00 +02:00
parent 7850bfc202
commit c2166b944f
4 changed files with 113 additions and 33 deletions
@@ -6,14 +6,14 @@ import de.jeanlucmakiola.agendula.data.tasks.room.TaskListEntity
import de.jeanlucmakiola.agendula.data.tasks.room.TasksDatabase
import de.jeanlucmakiola.caldav.CalDavDiscovery
import de.jeanlucmakiola.caldav.TaskCollection
import kotlinx.coroutines.CancellationException
import kotlinx.coroutines.CoroutineDispatcher
import kotlinx.coroutines.NonCancellable
import kotlinx.coroutines.flow.Flow
import kotlinx.coroutines.withContext
import kotlinx.coroutines.withTimeoutOrNull
import okhttp3.HttpUrl.Companion.toHttpUrlOrNull
import javax.inject.Inject
import javax.inject.Singleton
import kotlin.time.Duration.Companion.seconds
/**
* The one thing the sign-in flow needs from [AccountRepository].
@@ -235,22 +235,38 @@ class AccountRepository @Inject constructor(
withContext(io) {
syncTrigger.cancel(displayName)
revokeAppPassword(accountId)
// The lists survive as device-only lists, so their cursors must not:
// a re-added account would otherwise inherit a "reconciled recently"
// that was true of a different account's data.
cadence.forget(
database.taskLists().syncedForAccount(accountId).map { it.id }.toSet(),
)
// Play's Account Deletion policy does not apply to us — there is no
// Agendula account to delete — but "I want it gone from this device
// too" is a reasonable thing to want, and it is the only way to get
// the tasks off the device without also uninstalling.
if (deleteLocalData) database.taskLists().deleteForAccount(accountId)
accountState.setNeedsSignIn(accountId, false)
credentials.clear(accountId)
database.accounts().delete(accountId)
accounts.find(displayName)?.let { accounts.remove(it) }
// ⚠️ Uncancellable from here. Everything below is destructive and
// spread over four stores that cannot share a transaction, and the
// caller is a viewModelScope tied to the Settings destination — the
// user taps Remove, the screen slides away, and a couple of back
// gestures kill the scope mid-sequence. Only the DataStore writes can
// observe cancellation (every Room DAO here is blocking), so the
// realistic landing point is `cadence.forget`: the app password is
// already revoked server-side while the row survives holding it, and
// the account reads "sign in again" for a credential we ourselves
// invalidated. Land further in and the tasks are gone with the row
// still there. The tail is three DataStore writes, two deletes and an
// AccountManager call — bounded and sub-second, so finishing it is
// strictly better than stopping anywhere inside it.
withContext(NonCancellable) {
// The lists survive as device-only lists, so their cursors must
// not: a re-added account would otherwise inherit a "reconciled
// recently" that was true of a different account's data.
cadence.forget(
database.taskLists().syncedForAccount(accountId).map { it.id }.toSet(),
)
// Play's Account Deletion policy does not apply to us — there is
// no Agendula account to delete — but "I want it gone from this
// device too" is a reasonable thing to want, and it is the only
// way to get the tasks off the device without also uninstalling.
if (deleteLocalData) database.taskLists().deleteForAccount(accountId)
accountState.setNeedsSignIn(accountId, false)
credentials.clear(accountId)
database.accounts().delete(accountId)
accounts.find(displayName)?.let { accounts.remove(it) }
}
}
/**
@@ -266,17 +282,19 @@ class AccountRepository @Inject constructor(
val origin = account.principalUrl?.toHttpUrlOrNull() ?: return
val password = (credentials.get(accountId) as? CredentialStore.Secret.Present)?.value
?: return
runCatching {
// ⚠️ Bounded well below the shared client's 30 s connect / 120 s read.
// The user pressed "remove"; an unreachable server must not leave the
// account sitting on screen for half a minute with nothing happening.
// Losing the revocation is the lesser failure, and it is the one the
// user can still fix by hand on the server.
withTimeoutOrNull(REVOCATION_TIMEOUT) {
gateway.revokeAppPassword(
CalDavGateway.Credentials(username, password, origin),
)
}
// ⚠️ The budget lives on the request itself, in AppPassword.revoke.
// Wrapping this in withTimeoutOrNull only *looked* bounded: the call
// parks on a socket read that no cancellation can break, and withContext
// returns when its block does, so the deadline passed and we waited
// anyway — minutes, on a multi-homed host that stalls.
try {
gateway.revokeAppPassword(
CalDavGateway.Credentials(username, password, origin),
)
} catch (_: CancellationException) {
// Deliberately swallowed. If the caller went away mid-revoke we still
// want the removal to finish rather than stop half-done; the tail
// below runs uncancellable for the same reason.
}
}
@@ -284,7 +302,5 @@ class AccountRepository @Inject constructor(
/** M3 primary-ish blue; the user recolours a list from its own screen. */
const val DEFAULT_LIST_COLOR = 0xFF4C6FFF.toInt()
/** Long enough for a reachable server, short enough not to feel stuck. */
val REVOCATION_TIMEOUT = 5.seconds
}
}
@@ -4,6 +4,9 @@ import okhttp3.HttpUrl
import okhttp3.OkHttpClient
import okhttp3.Request
import java.io.IOException
import kotlin.time.Duration
import kotlin.time.Duration.Companion.seconds
import kotlin.time.toJavaDuration
/**
* Gives an app password back when the account is removed.
@@ -58,12 +61,33 @@ object AppPassword {
}
/**
* ⚠️ The budget is enforced here because nothing above can enforce it.
* `execute()` parks on a socket read, which no coroutine cancellation and no
* `Thread.interrupt` can break — only closing the socket does, which is what
* `callTimeout` does. Wrapping this call in `withTimeoutOrNull` instead
* returns only once the read has finished anyway, so the caller waits out
* the shared client's own budget: 30s *per resolved address*, doubled by the
* authenticator's retry, plus up to 120s of read timeout. Minutes, for a
* step whose own doc says it must not block the removal.
*
* Not set on the shared client on purpose: `callTimeout` bounds the whole
* exchange including the body, and a multiget of a large list over a slow
* link legitimately runs long.
*
* @return true when the server confirmed the revocation. False means the
* credential may still exist server-side — the caller carries on regardless.
*/
fun revoke(httpClient: OkHttpClient, principal: HttpUrl): Boolean = try {
fun revoke(
httpClient: OkHttpClient,
principal: HttpUrl,
timeout: Duration = REVOCATION_TIMEOUT,
): Boolean = try {
val url = ocsRootFor(principal).newBuilder().addPathSegments(PATH).build()
httpClient.newCall(
httpClient.newBuilder()
// Shares the pool and dispatcher, so this costs nothing.
.callTimeout(timeout.toJavaDuration())
.build()
.newCall(
Request.Builder()
.url(url)
.delete()
@@ -73,10 +97,14 @@ object AppPassword {
.header("OCS-APIRequest", "true")
.header("Accept", "application/json")
.build(),
).execute().use { it.isSuccessful }
).execute().use { it.isSuccessful }
} catch (_: IOException) {
// callTimeout throws InterruptedIOException, which lands here.
false
} catch (_: IllegalArgumentException) {
false
}
/** What a best-effort courtesy call is worth waiting for. */
val REVOCATION_TIMEOUT: Duration = 5.seconds
}
@@ -5,9 +5,13 @@ import okhttp3.HttpUrl.Companion.toHttpUrl
import okhttp3.OkHttpClient
import okhttp3.mockwebserver.MockResponse
import okhttp3.mockwebserver.MockWebServer
import okhttp3.mockwebserver.SocketPolicy
import org.junit.After
import org.junit.Before
import org.junit.Test
import kotlin.time.Duration.Companion.milliseconds
import kotlin.time.Duration.Companion.nanoseconds
import kotlin.time.Duration.Companion.seconds
class AppPasswordTest {
@@ -38,6 +42,30 @@ class AppPasswordTest {
.isEqualTo("https://baikal.example.com/")
}
@Test fun `a server that never answers does not hold the removal`() {
// Accepts the connection and says nothing — the off-VPN homelab shape.
server.enqueue(MockResponse().setSocketPolicy(SocketPolicy.NO_RESPONSE))
val startedAt = System.nanoTime()
val revoked = AppPassword.revoke(
httpClient,
server.url("/remote.php/dav/principals/users/alice/"),
timeout = 200.milliseconds,
)
val elapsed = (System.nanoTime() - startedAt).nanoseconds
// ⚠️ The budget has to live on the call. execute() parks on a socket read
// that neither coroutine cancellation nor Thread.interrupt can break, so
// a timeout imposed from outside returns only once the read has finished
// anyway — after the shared client's own minutes-long budget.
assertThat(revoked).isFalse()
assertThat(elapsed).isLessThan(2.seconds)
}
@Test fun `the default budget is what the removal is willing to wait`() {
assertThat(AppPassword.REVOCATION_TIMEOUT).isEqualTo(5.seconds)
}
@Test fun `the revocation is a DELETE with the OCS header`() {
server.enqueue(MockResponse().setResponseCode(200))
@@ -202,6 +202,14 @@ class CalDavHttpTest {
.header("WWW-Authenticate", "$scheme realm=\"dav\"")
.build()
@Test
fun `the shared client has no call timeout`() {
// ⚠️ Deliberate. A callTimeout bounds the whole exchange including the
// body, and a multiget of a large list over a slow link legitimately runs
// long. The one call that needs a budget sets its own.
assertThat(CalDavHttp.anonymous("Agendula").callTimeoutMillis).isEqualTo(0)
}
@Test
fun `derived clients share one connection pool`() {
// A fresh OkHttpClient per probe gives each its own pool and dispatcher