Trace Dependencies and Shared Failure Points
6 min read
Lesson objectives
- Apply trace dependencies and shared failure points in an operating case and justify the decision with evidence and limits.
Please sign in to use all the tools
Once you sign in, the reading tools are yours: larger or smaller text, typeface, line spacing, focus mode, highlighting and bold, and notes saved to your account and gathered in My study.
Three problems may appear at once: check-in slows, guests miss updates, and the rooms team cannot see its new task list. Treating them as three independent failures can scatter people across three workarounds while the shared cause, one network connection, goes untreated. A dependency map is not a search for technical blame. It shows how a service outcome is produced and what else fails when a resource is lost. Start with the guest outcomes already identified, then trace backward through operations to resources. Draw each arrow from a required resource toward the operation it sustains, and name the relationship: 'room allocation requires a current room-status record' means more than an unexplained line. One outcome can need several resources together. Restoration of one resource does not guarantee the outcome.
Divide nodes into four categories to expose items that an application list misses:
- physical utilities such as power or water
- information and systems such as the reservation record
- people and authority such as a supervisor who releases a room
- and external services such as transport or connectivity
A node need not fit only one category. What matters is an explicit substitution decision. For each node record who checks it, how its operation is observed at the point of service, and which substitute is permitted. If IT reports that the network is back, test booking verification, message delivery and rooms-team updates separately. The server may be reachable while one application shows stale data or a lobby terminal still cannot connect.
A common node can cause cascading effects. Do not sum the people on each branch as though they were distinct. The same guest may experience a delayed check-in and a missing update and thus appear twice. Track two measures: the number of critical services affected and the number of distinct guest parties affected. The first reveals operational spread; the second prevents inflation of human impact. Then add time. A standby generator may run but have limited fuel, or it may require a trained operator to start it. 'Generator present' is not a complete alternative. Ask what load it can serve, how long it lasts, who monitors it, and what happens if the operator cannot reach it. A capacity promise unsupported by an actual test should remain an assumption.
A simple comparison can consider how many critical outcomes depend on a node, how long its substitute lasts, how difficult it is to verify the node, and whether branches can be separated. Avoid multiplying subjective ratings into a number that resembles a scientific probability. If a resource supports both a guest-help path and room allocation, making the help path independent may be more urgent than speeding check-in, even where check-in volume is larger. There are two ways to address a common failure: strengthen the resource or remove a critical outcome's dependence on it. A separately tested guest-help phone may matter more during an outage than a standby server that does not cover room connections. The decision rests on the property's actual design and test results, not on the slogan that backup exists.
When comparing two substitutes, write a failure test for each. A mobile-data link might share a power source or provider route with the original. A printed arrivals sheet may be too old, visible to unauthorized people or hard to reconcile without duplicate assignments. 'Manual' is not a sufficient description. Specify throughput, who may use the route, when it stops, and how records will be corrected after the system returns. Some continuity errors arise after restoration rather than during the outage, when paper entries are duplicated or a room is released from an old status. Give every substitute a closing check and a responsible owner. If evidence is missing, mark it as untested instead of converting it into a reassuring green status.
Lesson summary
The worked map links the server to check-in and messages, and links power to the switchboard and the switchboard to a help call