The situation: 26 minutes with no trace on the server
An employee's watch, an element of the occupational safety system at the plant, may signal an alarm by vibration for 26 minutes while sending nothing to the server. For the person assessing the event, this looks like a gap in the record: it is known that the device was working, but not what its sensors decided, whether the alarm cycle was in progress, or whether the heartbeat was succeeding.
The problem is organizational as well as technical. After an incident the safety manager must write a report, and the IT analyst must answer whether the failure lay in the device, the connection or the software. Without a durable record, each of these hypotheses stays a guess, and the plant loses the chance to draw conclusions and name corrective actions.
Why the server alone was not enough
The server records what reaches it. When the device's connection is interrupted or disturbed, or when the watch itself holds back transmission, the server side goes silent, and silence is not proof of anything. A lack of heartbeats alone does not tell whether the device was switched off, out of range in a dead zone, or working and making decisions locally.
That is why the conclusion in this scenario is as follows: event records must be created where the events happen, that is on the device, and the server must store them durably. The approach is known from flight recorders in transport and industry: a black box does not depend on whether anyone is querying it at the moment.
- The server sees only the events that arrived; it does not see decisions taken locally.
- A gap in heartbeats can have many causes, and without a log they cannot be told apart.
- Evidence for a report should be durable, not reconstructed from the memory of the people present.
26-minute period in device cycles
The solution: an on-device log sent in batches
The watch keeps its own event log: sensor decisions, the alarm cycle and the results of successive heartbeats. The log is not sent entry by entry but in batches, every third heartbeat cycle. With a 26-second interval this gives a transmission roughly every 78 seconds, in batches of up to 89 entries. The queue on the device is limited to 89 entries, so the watch memory does not grow without end.
The server stores the received log durably. Two scenarios that spoil records in practice are also covered. The first is a repeated transmission after a lost acknowledgement: the uniqueness of an entry is enforced by the pair of user and identifier, so a repeated batch does not create duplicates. The second is a faulty entry within a batch: the server rejects the whole batch, so the device does not delete from its queue entries that the server did not store.
Two times in every entry
The log stores both the device time and the time of receipt by the server. The difference between them shows the length of the silence after the connection returns: if an entry was created on the watch and the server received it a dozen minutes later, it is known how long the gap in transmission lasted and when it ended.
How events are rebuilt step by step
For a 26-minute period the watch completes about 60 heartbeat cycles (1560 s divided by 26 s), and batch transmission corresponds to about 20 batches of 78 s. This is an example calculation based on a 26 s interval and transmission every third cycle; the actual number depends on the device configuration and the state of the connection.
With the log in hand, the analyst builds a timeline in four steps. First, a time window around the incident is selected based on the device time. Next, the entries are sorted and sensor decisions are separated from the alarm cycle and heartbeat results. Then the device time is compared with the time of receipt to establish when the connection returned. Finally, the timeline is matched against records from other systems, for example access control, and its consistency is checked.
- Selecting the time window according to the device time.
- Separating sensor decisions, the alarm cycle and heartbeat results.
- Calculating the silence from the difference between device time and receipt time.
- Matching against records from other systems and describing it in the report.
Limits of the solution to be aware of
The device queue is a window, not an archive. A watch offline for more than about 2 hours loses the start of its own failure, because the oldest entries are overwritten by newer ones. The practical rule follows: during long connection gaps the value of the log drops, and reconstructing the start of an event may become impossible.
The second limit concerns deployment. The logging function works on instances where the backend version supporting the log has been deployed. Before treating the log as evidence, it is worth checking that every instance receiving watches actually has this function.
Conclusions for the plant
Three conclusions follow from this case. First, the occupational safety system should record events on the device and not rely solely on what reaches the server. Second, batch transmission with protection against duplicates and against partial saves is safer than single-entry transmission. Third, two times in one entry are the cheapest way to measure a silence.
For the safety manager this means a real change: after an incident the report rests on a durable record, not on accounts. For the IT department it means a control task: verifying that all hosts have a log endpoint, and deciding how long the log is kept on the server.
How many cycles fit into 26 minutes of silence
| Parameter | Value |
|---|---|
| Silence duration | 26 min |
| Heartbeat interval | 26 s |
| Log transmission every | 3 cycles |
Result: About 60 heartbeat cycles and about 20 log batches in the 26-minute window.
Note: Example calculation based on the assumptions in the table; actual values depend on the device configuration and connection.
Video
Frequently asked questions
Does the watch log replace the server records?
No, it complements them. The server sees what arrived, while the log also shows decisions taken locally, including while the connection was interrupted.
What happens if a log batch is sent twice?
Nothing bad: the uniqueness of an entry is enforced by the pair of user and identifier, so the repetition does not create duplicates.
How long can the device be offline without losing the start of an event?
Roughly about 2 hours, because the queue is limited to 89 entries. After a longer gap the start of a failure may be overwritten by newer entries.
How does the analyst know how long the silence lasted?
From the difference between the device time and the time of receipt by the server, recorded in every log entry.
Want evidence after an incident?
Let's talk about how the device log fits into your event analysis process and the occupational safety report. We will show it in a demonstration.
Let's talk