Skip to content

Commit df4095b

Browse files
authored
Update azure.md for storage initializer environment variables (#541)
* Update azure.md Signed-off-by: Agoston Horvath <[email protected]> * Update azure.md Fix example Signed-off-by: Agoston Horvath <[email protected]> * Added AWS S3 doc Signed-off-by: Horvath Agoston <[email protected]> --------- Signed-off-by: Agoston Horvath <[email protected]> Signed-off-by: Horvath Agoston <[email protected]>
1 parent 070903c commit df4095b

File tree

2 files changed

+37
-0
lines changed

2 files changed

+37
-0
lines changed

docs/model-serving/storage/providers/azure.md

Lines changed: 19 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -97,6 +97,25 @@ Apply the `sklearn-azure.yaml`.
9797
kubectl apply -f sklearn-azure.yaml
9898
```
9999

100+
### Configuring blob download
101+
102+
There are a number of environment variables that allow to configure azure blob download parallelization:
103+
104+
```yaml
105+
kind: ClusterStorageContainer
106+
spec:
107+
container:
108+
env:
109+
- name: AZURE_MAX_FILE_CONCURRENCY
110+
value: 4
111+
- name: AZURE_MAX_CHUNK_CONCURRENCY
112+
value: 4
113+
```
114+
(only relevant entries shown)
115+
116+
The above example shows the default values. Together they control how many files are downloaded in parallel, and the download parallelism of each file.
117+
Note that each download has a buffer of 8MiB.
118+
100119
## Run a prediction
101120

102121
Now, the ingress can be accessed at `${INGRESS_HOST}:${INGRESS_PORT}` or follow [this instruction](../../../getting-started/predictive-first-isvc.md#4-determine-the-ingress-ip-and-ports)

docs/model-serving/storage/providers/s3/s3.md

Lines changed: 18 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -106,6 +106,24 @@ Apply the YAML file:
106106
kubectl apply -f mnist-s3.yaml
107107
```
108108

109+
### Configuring object download
110+
111+
There are a number of environment variables that allow to configure S3 object download parallelization:
112+
113+
```yaml
114+
kind: ClusterStorageContainer
115+
spec:
116+
container:
117+
env:
118+
- name: S3_MAX_FILE_CONCURRENCY
119+
value: 4
120+
```
121+
(only relevant entries shown)
122+
123+
The above example shows the default value. It controls how many files are downloaded in parallel from S3.
124+
Note that each parallel download forks python and creates a new boto3 client.
125+
126+
109127
## Run a prediction
110128
111129
Now, the ingress can be accessed at `${INGRESS_HOST}:${INGRESS_PORT}` or follow [this instruction](../../../../getting-started/predictive-first-isvc.md#4-determine-the-ingress-ip-and-ports)

0 commit comments

Comments
 (0)