When work piles up, making every stage faster is rarely the answer. The theory of constraints (ToC) focuses improvement on the stage that currently limits end-to-end throughput. In a development workflow, that might be code review; in a service architecture, it might be a shared database. Speeding up work elsewhere can simply enlarge the queue ahead of the constraint. section 6.4 section 11.4
Look for a necessary stage where demand persistently exceeds capacity, not just a task that takes a long time. Growing queues, longer waits, aging work items, and a low activity ratio are useful signals. If six developers open 15 pull requests a week but reviewers close only 10, five more requests wait each week. Faster coding or deployment will not clear that queue. section 6.4
Improve flow in this order:
- Identify the binding constraint using queues, capacity, and observed slowdowns. section 6.4
- Exploit its existing capacity: remove avoidable work or make better use of the capacity already available. section 6.4.1 section 11.4
- Subordinate upstream work to its pace. Limit work in progress or sequence integrations so the constraint is not overloaded. section 6.4.1
- Elevate its capacity if needed, then look again: removing one bottleneck can reveal the next. section 11.4
A shared database at peak demand
During a dinner rush, three services using the same PostgreSQL database slow sharply, while a payment service that does not use it barely changes. The shared database is the likely constraint. First improve its existing use, including queries and connection pooling. Then reduce demand through caching, batching, or moving suitable workloads elsewhere. If that is insufficient, consider read replicas, scaling, or splitting the database. Diagnose the next constraint after the change rather than assuming the system is finished. section 11.4
Table 6.1 Practical strategies for elevating the constraint
|
Type
|
Symptoms
|
Strategy
|
Example
|
|
Code review
|
Aging PRs, long review wait time
|
Reduce demand
|
Cap PR size at 200 lines
|
|
Code review
|
Reviewers spend too much time on formatting and flagging
internal standards
|
Remove trivial work
|
Automate linting, formatting
|
|
Code review
|
Limited reviewer bandwidth
|
Increase capacity
|
Dedicate a review block daily in the calendar
|
|
CI runtime
|
Long build queue
|
Parallelize
|
Split integration tests across runners
|
|
CI runtime
|
Flaky failures
|
Remove instability
|
Stabilize or remove flaky tests
|
|
Deployment approval
|
Manual gate delay
|
Replace gate with policy as code
|
Automated compliance checks
|
|
External dependency
|
API blocking progress
|
Decouple or mock
|
Contract testing with stubbed services
|
Compare symptoms with targeted interventions for review, CI, approvals, and external dependencies; the useful fix depends on which stage is actually binding. section 6.4.1
The practical question is not “What can we make faster?” but “What currently sets the pace for completed work?” Protect that stage from overload, improve it deliberately, and recheck the whole flow. section 6.4.1 section 11.4
