Add loadable checkpoint files for cc-metric-store

This commit is contained in:
Lou Knauer 2021-11-29 10:29:56 +01:00
parent 85bd467717
commit 5f4e218633
5 changed files with 41 additions and 23 deletions

1
.gitignore vendored
View File

@ -4,3 +4,4 @@ data/symfony
data/symfony/**
data/influxdb
data/sqldata
data/cc-metric-store

View File

@ -13,5 +13,7 @@ RUN cd /cc-metric-store && go build
RUN mv /cc-metric-store/cc-metric-store /go/bin
COPY config.json /go/bin
VOLUME /data
WORKDIR /go/bin
CMD ["./cc-metric-store"]

View File

@ -1,31 +1,29 @@
{
"metrics": {
"load_one": { "frequency": 10, "aggregation": null, "scope": "node" },
"load_five": { "frequency": 10, "aggregation": null, "scope": "node" },
"load_fifteen": { "frequency": 10, "aggregation": null, "scope": "node" },
"proc_run": { "frequency": 10, "aggregation": null, "scope": "node" },
"proc_total": { "frequency": 10, "aggregation": null, "scope": "node" },
"mem_free": { "frequency": 10, "aggregation": null, "scope": "node" },
"mem_used": { "frequency": 10, "aggregation": null, "scope": "node" },
"power": { "frequency": 10, "aggregation": "sum", "scope": "socket" },
"mem_bw": { "frequency": 10, "aggregation": "sum", "scope": "socket" },
"flops_sp": { "frequency": 10, "aggregation": "sum", "scope": "cpu" },
"flops_dp": { "frequency": 10, "aggregation": "sum", "scope": "cpu" },
"flops_any": { "frequency": 10, "aggregation": "sum", "scope": "cpu" },
"clock": { "frequency": 10, "aggregation": "avg", "scope": "cpu" },
"cpi": { "frequency": 10, "aggregation": "avg", "scope": "cpu" }
"clock": { "frequency": 60, "aggregation": null, "scope": "node" },
"cpi": { "frequency": 60, "aggregation": null, "scope": "node" },
"cpu_load": { "frequency": 60, "aggregation": null, "scope": "node" },
"flops_any": { "frequency": 60, "aggregation": null, "scope": "node" },
"flops_dp": { "frequency": 60, "aggregation": null, "scope": "node" },
"flops_sp": { "frequency": 60, "aggregation": null, "scope": "node" },
"ib_bw": { "frequency": 60, "aggregation": null, "scope": "node" },
"lustre_bw": { "frequency": 60, "aggregation": null, "scope": "node" },
"mem_bw": { "frequency": 60, "aggregation": null, "scope": "node" },
"mem_used": { "frequency": 60, "aggregation": null, "scope": "node" },
"rapl_power": { "frequency": 60, "aggregation": null, "scope": "node" }
},
"checkpoints": {
"interval": 21600,
"directory": "./var/checkpoints",
"restore": 43200
"interval": 100000000000,
"directory": "/data/checkpoints",
"restore": 100000000000
},
"archive": {
"interval": 86400,
"directory": "./var/archive"
"interval": 100000000000,
"directory": "/data/archive"
},
"retention-in-memory": 86400,
"retention-in-memory": 100000000000,
"http-api-address": "0.0.0.0:8081",
"nats": "nats://nats.backend:4222",
"nats": "nats://cc-nats:4222",
"jwt-public-key": "kzfYrYy+TzpanWZHJ5qSdMj5uKUWgq74BWhQG6copP0="
}

View File

@ -32,3 +32,16 @@ wget https://hpc-mover.rrze.uni-erlangen.de/HPC-Data/0x7b58aefb/eig7ahyo6fo2bais
cd influxdb/data
tar xJf ../../influxdbv2-data.tar.xz
rm ../../influxdbv2-data.tar.xz
cd ../..
# Download checkpoint files for cc-metric-store
mkdir -p cc-metric-store/checkpoints
mkdir -p cc-metric-store/archive
cd cc-metric-store/checkpoints
wget https://hpc-mover.rrze.uni-erlangen.de/HPC-Data/0x7b58aefb/eig7ahyo6fo2bais0ephuf2aitohv1ai/cc-metric-store.checkpoints-emmy.tar.gz
wget https://hpc-mover.rrze.uni-erlangen.de/HPC-Data/0x7b58aefb/eig7ahyo6fo2bais0ephuf2aitohv1ai/cc-metric-store.checkpoints-woody.tar.gz
tar xf cc-metric-store.checkpoints-emmy.tar.gz
tar xf cc-metric-store.checkpoints-woody.tar.gz
rm cc-metric-store.checkpoints-emmy.tar.gz cc-metric-store.checkpoints-woody.tar.gz
cd ../..

View File

@ -1,6 +1,6 @@
services:
nats:
container_name: nats
container_name: cc-nats
image: nats:alpine
ports:
- "4222:4222"
@ -40,9 +40,13 @@ services:
- ${DATADIR}/influxdb/config:/etc/influxdb2
cc-metric-store:
container_name: metric-store
container_name: cc-metric-store
build:
context: ./cc-metric-store
ports:
- "8081:8081"
volumes:
- ${DATADIR}/cc-metric-store:/data
depends_on:
- nats