F
Member
【Editorial Discussion】
An edge node that correctly stores data during a network outage still has to get that data into the historian when the link returns. A week of high-rate tags replayed at full speed can overload the server, scramble order, or hide a live outage under a backfill storm. “We have store-and-forward” is not the same as a tested replay policy.
1) Decide the order and the rate before the outage happens. Oldest-first versus newest-first, a bandwidth cap, and whether live data has priority over backfill should be written down. An unbounded replay will look like a historian incident.
2) Duplicate keys and overlapping timestamps need a rule. Replayed data that meets live data in the middle of the gap, or a node whose clock jumped, can create two values for one timestamp or an apparent step backwards. The identifier of the source and the quality or origin flag should survive the replay.
3) Not every tag deserves the same retention. Fast cyclic data, events, and alarms may have different gaps that are acceptable. A policy that stores everything at full rate for a week can fill the edge disk and still miss the event you cared about if the disk was already wrapping.
4) Test the failback, not only the failover. A planned disconnect of realistic length, followed by replay under production-like historian load, is the only way to know whether the policy works.
Open question: after a long outage, do you cap replay, drop the oldest, or stall live updates until the gap is closed? Have you filled a historian or an edge disk this way? Please share the policy that survived contact with a real gap.
An edge node that correctly stores data during a network outage still has to get that data into the historian when the link returns. A week of high-rate tags replayed at full speed can overload the server, scramble order, or hide a live outage under a backfill storm. “We have store-and-forward” is not the same as a tested replay policy.
1) Decide the order and the rate before the outage happens. Oldest-first versus newest-first, a bandwidth cap, and whether live data has priority over backfill should be written down. An unbounded replay will look like a historian incident.
2) Duplicate keys and overlapping timestamps need a rule. Replayed data that meets live data in the middle of the gap, or a node whose clock jumped, can create two values for one timestamp or an apparent step backwards. The identifier of the source and the quality or origin flag should survive the replay.
3) Not every tag deserves the same retention. Fast cyclic data, events, and alarms may have different gaps that are acceptable. A policy that stores everything at full rate for a week can fill the edge disk and still miss the event you cared about if the disk was already wrapping.
4) Test the failback, not only the failover. A planned disconnect of realistic length, followed by replay under production-like historian load, is the only way to know whether the policy works.
Open question: after a long outage, do you cap replay, drop the oldest, or stall live updates until the gap is closed? Have you filled a historian or an edge disk this way? Please share the policy that survived contact with a real gap.