Realm Mobile Client and Sync Server: Measuring Sync Latency Across Change Notification Boundaries

Question

We are evaluating sync latency between the Realm mobile client and Sync Server in current SDKs before optimization. The goal is to baseline how long a batch write takes to propagate and become observable through change notifications. Realm's change notification mechanism relies on JSON patch semantics, delivering diffs only when property values change, creating uncertainty about whether a client that re-queries after a batch write observes every intermediate state or only the final consolidated diff.

Additional constraints complicate measurement. The default sync endpoint enforces a 60-second session timeout, and prolonged offline periods require explicit re-authentication, adding variable delay. Cross-platform conflict resolution uses last-writer-wins timestamps, so concurrent edits can silently overwrite each other, complicating latency attribution. Realm Studio's timeline view and network logs are available, but it is unclear which signals correspond to server-side sync versus client-side notification batching.

We need to clarify the measurement boundary. Does the change notification mechanism guarantee that all property-level diffs from a batch write are delivered before the session timeout, or can a post-batch re-query miss intermediate states? When profiling sync latency across iOS and Android, should the session timeout and re-authentication delay be included in the latency budget, or treated as a separate metric? Is there a documented way to distinguish network-induced sync latency from local change-notification batching in Realm Studio's timeline view?

Tasadduq Burney · Answer

Answer
Realm change notifications deliver a single consolidated diff per transaction. A batch write does not produce multiple property-level notifications. A client that re-queries after the batch will observe only the final merged state, not intermediate states. The notification mechanism does not guarantee delivery of intermediate property-level diffs before session timeout.
Confirmed facts

One notification per transaction is delivered, containing a consolidated JSON patch diff.
Session timeout and re-authentication after prolonged offline periods are separate from core sync propagation. They should be measured separately unless the goal is full user-perceived round-trip latency including reconnect.
iOS and Android clients share the same notification behavior; platform latency differences come from OS scheduling and network stack, not Realm diff delivery guarantees.

Likely explanation
Network-induced sync latency and local notification batching are observable as distinct intervals in logs. The default 60-second session timeout can add variable delay if the client is idle, and last-writer-wins conflict resolution can silently drop concurrent updates, affecting perceived latency attribution.
Isolating network vs local batching in Realm Studio
In the timeline view:

Network latency is the gap between server Sent changeset and client Received changeset.
Local batching is the short interval between client Received changeset and the time the notification callback is invoked.

Steps for this case

Enable Realm Sync debug logging on server and client to capture Transaction committed, Changeset sent, Changeset received, Notification delivered timestamps.
Perform a controlled batch write while the client is online and connected. Record server commit time and client notification callback time.
Repeat after forcing session timeout >60s inactivity and after an offline period requiring re-auth. Compare added delay to baseline.
Verify batch size changes do not change the number of notification callbacks, confirming one consolidated diff per transaction.

Confirm whether your latency budget targets pure sync propagation or full round-trip including re-authentication, as that changes whether timeout delay is included.

Realm Mobile Client and Sync Server: Measuring Sync Latency Across Change Notification Boundaries

1 answer

Answer

Confirmed facts

Likely explanation

Isolating network vs local batching in Realm Studio

Steps for this case

Your perspective belongs here.

0 question comments