Fluksio/app
0
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

GPUs follow the engine's own CUDA_VISIBLE_DEVICES

Browse Source 
Its entries are the cards present, and a local child's allocation is
named through them, so under =2,3 card 0 is device 2. Remote workers
keep mapping against their own mask.
This commit is contained in:
stroblme
2026-10-15 09:53:38 +02:00
parent 290b9fa705
commit f2a823218a
5 changed files with 69 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docs/code/cli.md
+3 -2
View File
@@ -65,8 +65,9 @@ that instance's running engine; plain mode refuses to start a duplicate.
| `--worker-max-rss MB` | `0` (disabled) | Retire a worker above this resident memory after its call |
Resource defaults can also come from the corresponding `FLOW_*` settings;
see [Configuration](../reference/configuration.md). GPU detection uses
NVIDIA device nodes. Set `--gpus` when your hardware is not detected.
see [Configuration](../reference/configuration.md). GPU detection counts the
entries of `CUDA_VISIBLE_DEVICES` when it is set, otherwise NVIDIA device
nodes. Set `--gpus` when your hardware is not detected.
Use local disk for the data directory on a cluster: SQLite's write-ahead log
does not support NFS.
docs/reference/configuration.md
+7 -3
View File
@@ -126,7 +126,7 @@ Set `TZ` in `.env` for schedules that should follow local time, such as
| `FLOW_MAX_RUNS` | `4` | concurrent batch runs |
| `FLOW_NODE_TIMEOUT` | `0` | maximum node silence in seconds, unless the node overrides it; `0` disables |
| `FLOW_CPUS` | `0` | CPU budget; `0` detects cores and reserves two for the engine |
| `FLOW_GPUS` | `0` | local GPU count; `0` detects NVIDIA device nodes |
| `FLOW_GPUS` | `0` | local GPU count; `0` counts the engine's `CUDA_VISIBLE_DEVICES`, or NVIDIA device nodes without it |
| `FLOW_RAM` | `0` | memory budget in MB; `0` detects host/container memory and reserves 1 GiB |
| `FLOW_WORKER_MAX_RSS_MB` | `0` | retire a local worker above this memory use after a call; `0` disables |
| `FLOW_NODE_STOP_TIMEOUT` | `5` | seconds a node gets to stop during a rebuild before the rebuild moves on |
@@ -139,14 +139,18 @@ Set `TZ` in `.env` for schedules that should follow local time, such as
The three concurrency limits must be at least `1`. They also have CLI flags:
`--max-workers`, `--max-cascades` and `--max-runs`. When `FLOW_MAX_WORKERS` or
`FLOW_MAX_CASCADES` is more than four times the machine's core count, the
engine logs a warning at startup and uses the value anyway. Resource flags are `--gpus`,
`--ram` and `--worker-max-rss`.
engine logs a warning at startup and uses the value anyway. Resource flags are
`--gpus`, `--ram` and `--worker-max-rss`.
Nodes with declared resources are scheduled against the available budget.
Undeclared nodes use the shared pool and receive thread caps. Attached workers
report their own inventory; the engine's resource settings describe only its
own machine. See [Remote workers](../code/workers.md) and the **Workers** screen.
When the engine runs with `CUDA_VISIBLE_DEVICES`, a node's card is named the
way that list names it, so under `2,3` the first card is device `2`. UUID and
MIG entries are passed on as written.
## Artifacts
| Variable | Default | Purpose |