Skip to content

feat(process): processMemory.summarize 汇总压缩/换出内存、GPU 显存与 CPU、GPU 占用 #45

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
xiaoyumuxi merged 1 commit into from
Nov 14, 2026
Merged

feat(process): processMemory.summarize 汇总压缩/换出内存、GPU 显存与 CPU、GPU 占用 #45

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
68 changes: 54 additions & 14 deletions rust/contracts/rust-core-api.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -550,7 +550,7 @@ or removing a command fails until the table changes in the same commit.
| `processLifecycle.resolve` | Decide how each process or resource of one Lithe instance ends |
| `processLifecycle.inventory` | Group a process snapshot of one Lithe instance into terminals, Agent tasks and background services |
| `processLifecycle.settle` | Turn confirmation answers into what ends, what is kept and the updated settings |
| `processMemory.summarize` | Sum a process table's memory by part of one Lithe instance: the host, language servers, plugin services, system helpers and other child processes |
| `processMemory.summarize` | Sum a process table's memory, compressed memory, graphics memory and processor and graphics use by part of one Lithe instance: the host, language servers, plugin services, system helpers and other child processes |
| `markdown.render` | Renders and sanitizes shared Markdown |
| `markdown.outline` | Outlines the headings of shared Markdown as a document-symbol tree |
| `markdown.resolveReference` | Decides what a link or an image of a rendered Markdown document points at |
Expand Down Expand Up @@ -5080,12 +5080,21 @@ scope-token ownership. Fixtures live in

Fixtures live in `rust/fixtures/process-lifecycle/settle-v1.json`.

`processMemory.summarize` accepts `{ hostProcessId, processes, roles, entryLimit }`
and attributes every process the host could read to one part of the instance,
so every host shows the same numbers for the same table. Each process row is
`{ processId, parentProcessId, responsibleProcessId?, startTime?, name, bytes }`.
`bytes` is the memory the host charges to the process; every row of one request
uses the same measure. `responsibleProcessId` is the process the operating system
`processMemory.summarize` accepts `{ hostProcessId, processes, roles, entryLimit,
sampledAt?, previous?, logicalProcessors? }` and attributes every process the host
could read to one part of the instance, so every host shows the same numbers for
the same table. Each process row is `{ processId, parentProcessId,
responsibleProcessId?, startTime?, name, bytes, compressedBytes?, gpuMemoryBytes?,
cpuTime?, gpuEngines? }`. Every host reports memory the same way: `bytes` is
what is in memory, and `compressedBytes` is what the system compressed or wrote
out to disk, which `bytes` does not include. `gpuMemoryBytes` is graphics memory
charged to the process that `bytes` does not hold, such as dedicated video memory
on Windows; a host whose `bytes` already holds it leaves it out. `cpuTime` is the
processor time the process used since it started, user and system together, in
microseconds. `gpuEngines` are `{ engine, busyTime }`: the microseconds each
graphics engine was busy with the process's work, where `engine` names the same
engine for every process and table of one host and times of one name add up.
`responsibleProcessId` is the process the operating system
holds responsible, which macOS reports for XPC services. `startTime` is in
microseconds since 1970; a reported parent that started after its child is a
reused process ID and is not followed. `roles` are
Expand All @@ -5109,14 +5118,45 @@ Every process counts at most once:
- Other processes are left out. The processes of a cycle of parent links are
decided as if none of them had a parent.

It returns `{ totalBytes, categories, entries, omitted }`. `categories` lists
It returns `{ totalBytes, compressedBytes?, gpuMemoryBytes?, cpuPercent?,
gpuPercent?, categories, entries, omitted, baseline? }`. `categories` lists
`host`, `languageServer`, `pluginService`, `systemHelper` and `childProcess` in
that order, each `{ kind, bytes, processCount }`, empty ones included. `entries`
are the groups other than the host, `{ kind, name, processId, bytes,
processCount }`, largest first, with ties ordered by part, name and process ID.
`name` is the role name, else the process name, shortened to 80 characters.
`omitted` is `{ count, bytes }` of the groups past the limit. A table without the
host or with a repeated process ID is `invalid_request`. Fixtures live in
that order, each `{ kind, bytes, processCount }` with the same optional figures,
empty ones included. `entries` are the groups other than the host, `{ kind, name,
processId, bytes, processCount }` with the same optional figures, largest first by
`bytes`, with ties ordered by part, name and process ID. `name` is the role name,
else the process name, shortened to 80 characters. `omitted` is `{ count, bytes }`
of the groups past the limit.

Each optional figure is on the summary, every category and every entry, or on
none of them. `compressedBytes` and `gpuMemoryBytes` are there when a process of
the instance reported them and sum the processes that did.

Percentages measure use between two tables of one host:

- `sampledAt` is when the table was read, in microseconds of a clock that only
moves forward. A request with it gets `baseline: { sampledAt, processes }`: the
instance's processes that have `cpuTime` or `gpuEngines`, by process ID, each
`{ processId, startTime?, cpuTime?, gpuEngines? }` with one engine row per name.
The host sends it back as `previous` with its next table.
- A process is compared with the `previous` row of its process ID and start time;
a row with another start time belongs to an earlier process with the same ID.
A process the baseline does not have, such as one started since, adds nothing
until the next table. A counter lower than its baseline adds nothing.
- `cpuPercent` is there when a process of the instance has `cpuTime` in both
tables. It is the processor time used over the interval, where 100 is one
processor busy, as Activity Monitor and `top` show it. With
`logicalProcessors` it is a share of that many processors instead, as the
Windows Task Manager shows it, and at most 100.
- `gpuPercent` is there when a process of the instance has engines in both
tables. The busy times of each engine are added up over the group, and the
busiest engine's share of the interval is the percentage, at most 100. An engine
the baseline row does not list counts from zero.
- Percentages are rounded to a tenth. Without `sampledAt`, without `previous`, or
with a `sampledAt` no later than the baseline's, there are none.

A table without the host or with a repeated process ID, or a `logicalProcessors`
of 0, is `invalid_request`. Fixtures live in
`rust/fixtures/process-lifecycle/memory-summary-v1.json`.

## Local IDE capability broker
Expand Down
Loading
Loading