Skip to content
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

Jump to bottom

Use jumpbox runners for MGMN/SLURM jobs #1103

Open
wants to merge 6 commits into
base: main
Choose a base branch
from
+102 −111
Open

Use jumpbox runners for MGMN/SLURM jobs #1103

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
d0800ed
make all SLURM job run on jumpbox runners
yhtang Oct 15, 2024
90a50db
limit max parallelism
yhtang Oct 15, 2024
743381e
sandbox
yhtang Oct 15, 2024
8485869
sandbox
yhtang Oct 15, 2024
63b692d
use standard SSH port and IP whitelist for security
yhtang Oct 15, 2024
4e955e7
reset sandbox
yhtang Oct 15, 2024
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
6 changes: 3 additions & 3 deletions .github/workflows/_runner_ondemand_slurm.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -20,7 +20,7 @@ on:
jobs:

launch-slurm-runner:
runs-on: ubuntu-latest
runs-on: jumpbox
steps:
- name: Print environment variables
run: env
Expand Down Expand Up @@ -58,7 +58,7 @@ jobs:
shell: bash -x -e {0}
run: |
SLURM_JOB_ID_FILE=$(mktemp)
ssh -p 3000 ${{ secrets.CLUSTER_LOGIN_USER }}@${{ vars.HOSTNAME_SLURM_LOGIN }} >${SLURM_JOB_ID_FILE} \
ssh ${{ secrets.CLUSTER_LOGIN_USER }}@${{ vars.HOSTNAME_SLURM_LOGIN }} >${SLURM_JOB_ID_FILE} \
sbatch --parsable \
<<"EOF"
#!/bin/bash
Expand Down Expand Up @@ -117,5 +117,5 @@ jobs:
if: cancelled()
shell: bash -x -e {0}
run: |
ssh -p 3000 ${{ secrets.CLUSTER_LOGIN_USER }}@${{ vars.HOSTNAME_SLURM_LOGIN }} \
ssh ${{ secrets.CLUSTER_LOGIN_USER }}@${{ vars.HOSTNAME_SLURM_LOGIN }} \
scancel ${{ steps.submit.outputs.SLURM_JOB_ID }}
26 changes: 12 additions & 14 deletions .github/workflows/_test_maxtext.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -37,14 +37,13 @@ jobs:

single-process-multi-device:
strategy:
max-parallel: 1
matrix:
PARALLEL_CONFIG:
- [1, 1, 2, 4]
# - [1, 1, 1, 8] # PP, DP, FSDP, TP
fail-fast: false

runs-on: ubuntu-22.04

runs-on: jumpbox
steps:
- name: Print environment variables
run: env
Expand Down Expand Up @@ -88,7 +87,7 @@ jobs:
id: submit
shell: bash -O expand_aliases -x -e {0}
run: |
alias sshx='ssh -p 3000 -o "ServerAliveInterval 7" ${{ secrets.CLUSTER_LOGIN_USER }}@${{ vars.HOSTNAME_SLURM_LOGIN }}'
alias sshx='ssh -o "ServerAliveInterval 7" ${{ secrets.CLUSTER_LOGIN_USER }}@${{ vars.HOSTNAME_SLURM_LOGIN }}'
sshx "date && hostname && sinfo"
sshx mkdir -p ${{ steps.meta.outputs.MODEL_PATH }}
JOB=$(sshx sbatch --parsable << EOF
Expand Down Expand Up @@ -149,17 +148,17 @@ jobs:
if: cancelled()
shell: bash -x -e {0}
run: |
ssh -p 3000 ${{ secrets.CLUSTER_LOGIN_USER }}@${{ vars.HOSTNAME_SLURM_LOGIN }} \
ssh ${{ secrets.CLUSTER_LOGIN_USER }}@${{ vars.HOSTNAME_SLURM_LOGIN }} \
scancel ${{ steps.submit.outputs.SLURM_JOB_ID }}

- name: Retrieve training logs and upload to TensorBoard server
shell: bash -x -e {0}
run: |
mkdir output/
rsync -rtz --progress -e 'ssh -p 3000' \
rsync -rtz --progress\
${{ secrets.CLUSTER_LOGIN_USER }}@${{ vars.HOSTNAME_SLURM_LOGIN }}:${{ steps.meta.outputs.LOG_FILE }} \
output/${{ steps.meta.outputs.TEST_CASE_NAME }}.log || true
rsync -rtz --progress -e 'ssh -p 3000' \
rsync -rtz --progress\
${{ secrets.CLUSTER_LOGIN_USER }}@${{ vars.HOSTNAME_SLURM_LOGIN }}:${{ steps.meta.outputs.MODEL_PATH }}/* \
output/ || true
rsync -rtz --progress \
Expand All @@ -184,6 +183,7 @@ jobs:

maxtext-multinode:
strategy:
max-parallel: 1
matrix:
PARALLEL_CONFIG:
- [1, 1, 1, 1]
Expand All @@ -193,9 +193,7 @@ jobs:
- [1, 2, 2, 2]
- [1, 4, 2, 2]
fail-fast: false

runs-on: ubuntu-22.04

runs-on: jumpbox
steps:
- name: Print environment variables
run: env
Expand Down Expand Up @@ -240,7 +238,7 @@ jobs:
id: submit
shell: bash -O expand_aliases -x -e {0}
run: |
alias sshx='ssh -p 3000 -o "ServerAliveInterval 7" ${{ secrets.CLUSTER_LOGIN_USER }}@${{ vars.HOSTNAME_SLURM_LOGIN }}'
alias sshx='ssh -o "ServerAliveInterval 7" ${{ secrets.CLUSTER_LOGIN_USER }}@${{ vars.HOSTNAME_SLURM_LOGIN }}'
sshx "date && hostname && sinfo"
sshx mkdir -p ${{ steps.meta.outputs.MODEL_PATH }}
JOB=$(sshx sbatch --parsable << EOF
Expand Down Expand Up @@ -304,17 +302,17 @@ jobs:
if: cancelled()
shell: bash -x -e {0}
run: |
ssh -p 3000 ${{ secrets.CLUSTER_LOGIN_USER }}@${{ vars.HOSTNAME_SLURM_LOGIN }} \
ssh ${{ secrets.CLUSTER_LOGIN_USER }}@${{ vars.HOSTNAME_SLURM_LOGIN }} \
scancel ${{ steps.submit.outputs.SLURM_JOB_ID }}

- name: Retrieve training logs and upload to TensorBoard server
shell: bash -x -e {0}
run: |
mkdir output/
rsync -rtz --progress -e 'ssh -p 3000' \
rsync -rtz --progress\
${{ secrets.CLUSTER_LOGIN_USER }}@${{ vars.HOSTNAME_SLURM_LOGIN }}:${{ steps.meta.outputs.LOG_FILE }} \
output/${{ steps.meta.outputs.TEST_CASE_NAME }}.log || true
rsync -rtz --progress -e 'ssh -p 3000' \
rsync -rtz --progress\
${{ secrets.CLUSTER_LOGIN_USER }}@${{ vars.HOSTNAME_SLURM_LOGIN }}:${{ steps.meta.outputs.MODEL_PATH }}/* \
output/ || true
rsync -rtz --progress \
Expand Down
62 changes: 30 additions & 32 deletions .github/workflows/_test_pax_rosetta.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -32,17 +32,15 @@ jobs:

single-process-multi-device-te:
strategy:
max-parallel: 1
matrix:
PARALLEL_CONFIG:
- [1, 8, 1, 1]
- [1, 1, 2, 4]
fail-fast: false

runs-on: ubuntu-22.04

runs-on: jumpbox
env:
BADGE_FILENAME_PREFIX: badge-rosetta-pax-single-process-multi-device-te

steps:
- name: Print environment variables
run: env
Expand Down Expand Up @@ -87,7 +85,7 @@ jobs:
shell: bash -O expand_aliases -x -e {0}
run: |
cd $GITHUB_WORKSPACE
alias sshx='ssh -p 3000 -o "ServerAliveInterval 7" ${{ secrets.CLUSTER_LOGIN_USER }}@${{ vars.HOSTNAME_SLURM_LOGIN }}'
alias sshx='ssh -o "ServerAliveInterval 7" ${{ secrets.CLUSTER_LOGIN_USER }}@${{ vars.HOSTNAME_SLURM_LOGIN }}'
sshx "date && hostname && sinfo"
sshx mkdir -p ${{ steps.meta.outputs.MODEL_PATH }}
JOB=$(sshx sbatch --parsable << EOF
Expand Down Expand Up @@ -146,18 +144,18 @@ jobs:
if: cancelled()
shell: bash -x -e {0}
run: |
ssh -p 3000 ${{ secrets.CLUSTER_LOGIN_USER }}@${{ vars.HOSTNAME_SLURM_LOGIN }} \
ssh ${{ secrets.CLUSTER_LOGIN_USER }}@${{ vars.HOSTNAME_SLURM_LOGIN }} \
scancel ${{ steps.submit.outputs.SLURM_JOB_ID }}

- name: Retrieve training logs and upload to TensorBoard server
shell: bash -x -e {0}
run: |
cd $GITHUB_WORKSPACE
mkdir output/
rsync -rtz --progress -e 'ssh -p 3000' \
rsync -rtz --progress\
${{ secrets.CLUSTER_LOGIN_USER }}@${{ vars.HOSTNAME_SLURM_LOGIN }}:${{ steps.meta.outputs.LOG_FILE }} \
output/${{ steps.meta.outputs.TEST_CASE_NAME }}.log || true
rsync -rtz --progress -e 'ssh -p 3000' \
rsync -rtz --progress\
${{ secrets.CLUSTER_LOGIN_USER }}@${{ vars.HOSTNAME_SLURM_LOGIN }}:${{ steps.meta.outputs.MODEL_PATH }}/* \
output/ || true
rsync -rtz --progress \
Expand Down Expand Up @@ -223,6 +221,7 @@ jobs:

rosetta-pax-multi-node-te:
strategy:
max-parallel: 1
matrix:
include:
- TEST_NAME: 1DP1FSDP1TP1PP_TE
Expand Down Expand Up @@ -259,8 +258,7 @@ jobs:
EVALUATE: true
ADDITIONAL_ARGS: "--model-type LLaMA70BProxy --evaluate"
fail-fast: false

runs-on: ubuntu-22.04
runs-on: jumpbox
env:
BADGE_FILENAME_PREFIX: badge-rosetta-pax-multi-node-te
steps:
Expand Down Expand Up @@ -308,7 +306,7 @@ jobs:
id: submit
shell: bash -O expand_aliases -x -e {0}
run: |
alias sshx='ssh -p 3000 -o "ServerAliveInterval 7" ${{ secrets.CLUSTER_LOGIN_USER }}@${{ vars.HOSTNAME_SLURM_LOGIN }}'
alias sshx='ssh -o "ServerAliveInterval 7" ${{ secrets.CLUSTER_LOGIN_USER }}@${{ vars.HOSTNAME_SLURM_LOGIN }}'
sshx "date && hostname && sinfo"
sshx mkdir -p ${{ steps.meta.outputs.MODEL_PATH }}
JOB=$(sshx sbatch --parsable << EOF
Expand Down Expand Up @@ -372,18 +370,18 @@ jobs:
if: cancelled()
shell: bash -x -e {0}
run: |
ssh -p 3000 ${{ secrets.CLUSTER_LOGIN_USER }}@${{ vars.HOSTNAME_SLURM_LOGIN }} \
ssh ${{ secrets.CLUSTER_LOGIN_USER }}@${{ vars.HOSTNAME_SLURM_LOGIN }} \
scancel ${{ steps.submit.outputs.SLURM_JOB_ID }}

- name: Retrieve training logs and upload to TensorBoard server
shell: bash -x -e {0}
run: |
cd $GITHUB_WORKSPACE
mkdir output/
rsync -rtz --progress -e 'ssh -p 3000' \
rsync -rtz --progress\
${{ secrets.CLUSTER_LOGIN_USER }}@${{ vars.HOSTNAME_SLURM_LOGIN }}:${{ steps.meta.outputs.LOG_FILE }} \
output/${{ steps.meta.outputs.TEST_CASE_NAME }}.log || true
rsync -rtz --progress -e 'ssh -p 3000' \
rsync -rtz --progress\
${{ secrets.CLUSTER_LOGIN_USER }}@${{ vars.HOSTNAME_SLURM_LOGIN }}:${{ steps.meta.outputs.MODEL_PATH }}/* \
output/ || true
rsync -rtz --progress \
Expand Down Expand Up @@ -450,15 +448,15 @@ jobs:

rosetta-pax-multi-node:
strategy:
max-parallel: 1
matrix:
PARALLEL_CONFIG:
- [1, 8, 1, 1]
- [1, 4, 1, 2]
- [4, 2, 1, 1]
- [4, 2, 1, 2]
fail-fast: false

runs-on: ubuntu-22.04
runs-on: jumpbox
env:
BADGE_FILENAME_PREFIX: badge-rosetta-pax-multi-node
steps:
Expand Down Expand Up @@ -506,7 +504,7 @@ jobs:
shell: bash -O expand_aliases -x -e {0}
run: |
cd $GITHUB_WORKSPACE
alias sshx='ssh -p 3000 -o "ServerAliveInterval 7" ${{ secrets.CLUSTER_LOGIN_USER }}@${{ vars.HOSTNAME_SLURM_LOGIN }}'
alias sshx='ssh -o "ServerAliveInterval 7" ${{ secrets.CLUSTER_LOGIN_USER }}@${{ vars.HOSTNAME_SLURM_LOGIN }}'
sshx "date && hostname && sinfo"
sshx mkdir -p ${{ steps.meta.outputs.MODEL_PATH }}
JOB=$(sshx sbatch --parsable << EOF
Expand Down Expand Up @@ -567,18 +565,18 @@ jobs:
if: cancelled()
shell: bash -x -e {0}
run: |
ssh -p 3000 ${{ secrets.CLUSTER_LOGIN_USER }}@${{ vars.HOSTNAME_SLURM_LOGIN }} \
ssh ${{ secrets.CLUSTER_LOGIN_USER }}@${{ vars.HOSTNAME_SLURM_LOGIN }} \
scancel ${{ steps.submit.outputs.SLURM_JOB_ID }}

- name: Retrieve training logs and upload to TensorBoard server
shell: bash -x -e {0}
run: |
cd $GITHUB_WORKSPACE
mkdir output/
rsync -rtz --progress -e 'ssh -p 3000' \
rsync -rtz --progress\
${{ secrets.CLUSTER_LOGIN_USER }}@${{ vars.HOSTNAME_SLURM_LOGIN }}:${{ steps.meta.outputs.LOG_FILE }} \
output/${{ steps.meta.outputs.TEST_CASE_NAME }}.log || true
rsync -rtz --progress -e 'ssh -p 3000' \
rsync -rtz --progress\
${{ secrets.CLUSTER_LOGIN_USER }}@${{ vars.HOSTNAME_SLURM_LOGIN }}:${{ steps.meta.outputs.MODEL_PATH }}/* \
output/ || true
rsync -rtz --progress \
Expand Down Expand Up @@ -645,12 +643,12 @@ jobs:

rosetta-pax-single-node-dropout-te:
strategy:
max-parallel: 1
matrix:
PARALLEL_CONFIG:
- [1, 8, 1, 1]
fail-fast: false

runs-on: ubuntu-22.04
runs-on: jumpbox
env:
BADGE_FILENAME_PREFIX: badge-rosetta-pax-single-node-dropout-te
steps:
Expand Down Expand Up @@ -698,7 +696,7 @@ jobs:
id: submit
shell: bash -O expand_aliases -x -e {0}
run: |
alias sshx='ssh -p 3000 -o "ServerAliveInterval 7" ${{ secrets.CLUSTER_LOGIN_USER }}@${{ vars.HOSTNAME_SLURM_LOGIN }}'
alias sshx='ssh -o "ServerAliveInterval 7" ${{ secrets.CLUSTER_LOGIN_USER }}@${{ vars.HOSTNAME_SLURM_LOGIN }}'
sshx "date && hostname && sinfo"
sshx mkdir -p ${{ steps.meta.outputs.MODEL_PATH }}
JOB=$(sshx sbatch --parsable << EOF
Expand Down Expand Up @@ -762,18 +760,18 @@ jobs:
if: cancelled()
shell: bash -x -e {0}
run: |
ssh -p 3000 ${{ secrets.CLUSTER_LOGIN_USER }}@${{ vars.HOSTNAME_SLURM_LOGIN }} \
ssh ${{ secrets.CLUSTER_LOGIN_USER }}@${{ vars.HOSTNAME_SLURM_LOGIN }} \
scancel ${{ steps.submit.outputs.SLURM_JOB_ID }}

- name: Retrieve training logs and upload to TensorBoard server
shell: bash -x -e {0}
run: |
cd $GITHUB_WORKSPACE
mkdir output/
rsync -rtz --progress -e 'ssh -p 3000' \
rsync -rtz --progress\
${{ secrets.CLUSTER_LOGIN_USER }}@${{ vars.HOSTNAME_SLURM_LOGIN }}:${{ steps.meta.outputs.LOG_FILE }} \
output/${{ steps.meta.outputs.TEST_CASE_NAME }}.log || true
rsync -rtz --progress -e 'ssh -p 3000' \
rsync -rtz --progress\
${{ secrets.CLUSTER_LOGIN_USER }}@${{ vars.HOSTNAME_SLURM_LOGIN }}:${{ steps.meta.outputs.MODEL_PATH }}/* \
output/ || true
rsync -rtz --progress \
Expand Down Expand Up @@ -839,12 +837,12 @@ jobs:

single-process-evaluation-te:
strategy:
max-parallel: 1
matrix:
PARALLEL_CONFIG:
- [1, 8, 1, 1]
fail-fast: false

runs-on: ubuntu-22.04
runs-on: jumpbox
env:
BADGE_FILENAME_PREFIX: badge-rosetta-pax-single-process-evaluation-te
steps:
Expand Down Expand Up @@ -890,7 +888,7 @@ jobs:
shell: bash -O expand_aliases -x -e {0}
run: |
cd $GITHUB_WORKSPACE
alias sshx='ssh -p 3000 -o "ServerAliveInterval 7" ${{ secrets.CLUSTER_LOGIN_USER }}@${{ vars.HOSTNAME_SLURM_LOGIN }}'
alias sshx='ssh -o "ServerAliveInterval 7" ${{ secrets.CLUSTER_LOGIN_USER }}@${{ vars.HOSTNAME_SLURM_LOGIN }}'
sshx "date && hostname && sinfo"
sshx mkdir -p ${{ steps.meta.outputs.MODEL_PATH }}
JOB=$(sshx sbatch --parsable << EOF
Expand Down Expand Up @@ -952,18 +950,18 @@ jobs:
if: cancelled()
shell: bash -x -e {0}
run: |
ssh -p 3000 ${{ secrets.CLUSTER_LOGIN_USER }}@${{ vars.HOSTNAME_SLURM_LOGIN }} \
ssh ${{ secrets.CLUSTER_LOGIN_USER }}@${{ vars.HOSTNAME_SLURM_LOGIN }} \
scancel ${{ steps.submit.outputs.SLURM_JOB_ID }}

- name: Retrieve training logs and upload to TensorBoard server
shell: bash -x -e {0}
run: |
cd $GITHUB_WORKSPACE
mkdir output/
rsync -rtz --progress -e 'ssh -p 3000' \
rsync -rtz --progress\
${{ secrets.CLUSTER_LOGIN_USER }}@${{ vars.HOSTNAME_SLURM_LOGIN }}:${{ steps.meta.outputs.LOG_FILE }} \
output/${{ steps.meta.outputs.TEST_CASE_NAME }}.log || true
rsync -rtz --progress -e 'ssh -p 3000' \
rsync -rtz --progress\
${{ secrets.CLUSTER_LOGIN_USER }}@${{ vars.HOSTNAME_SLURM_LOGIN }}:${{ steps.meta.outputs.MODEL_PATH }}/* \
output/ || true
rsync -rtz --progress \
Expand Down
Loading
Loading