Decoding the Two Lines
The most confusing part of PSI for newcomers is the difference between the some and full lines. Understanding this distinction is what separates someone who can read PSI from someone who truly understands it.
some — Impact on Latency
some pressure means: "At least one task was stalled waiting for this resource during this time window, even if other tasks were still making progress."
Think of a restaurant kitchen with four chefs. If one chef is blocked waiting for an ingredient from the walk-in freezer while the other three are still cooking, the kitchen is still producing food — but it's slower than it could be. That's some pressure. The resource is under contention. Latency is increasing, but throughput hasn't collapsed.
In practice: some CPU pressure of 20% means that for 20% of the last 10 seconds, at least one runnable task was sitting in the scheduler queue waiting for a core. Your p99 latency is suffering.
full — Impact on Throughput
full pressure means: "Every single runnable, non-idle task was simultaneously stalled. No useful user-space work was being done at all."
Back to the kitchen: full pressure is when all four chefs are blocked waiting for ingredients. The kitchen has ground to a halt. The hardware (the stoves, the counters) is sitting idle not because there's nothing to cook, but because everyone is stuck waiting for something else.
full pressure is the more alarming metric. It directly measures lost throughput — wall-clock time during which your application made zero progress.
The CPU exception: The /proc/pressure/cpu file only has a some line, not a full line. This is mathematically intentional: if tasks are waiting for CPU, the CPU must be running something (the task that got the core). It is impossible for all tasks to be simultaneously waiting for CPU while the CPU sits completely idle. Memory and I/O don't have this constraint — the disk or RAM bus can be saturated while every process is blocked waiting for it.