update observability and dashboards (#198)

This commit is contained in:
Adil Hafeez 2024-10-18 15:07:49 -07:00 committed by GitHub
parent f189d5703b
commit faf64960df
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
6 changed files with 174 additions and 166 deletions

View file

@ -115,26 +115,20 @@ static_resources:
domains:
- "*"
routes:
{% for internal_clustrer in ["embeddings", "zeroshot", "guard", "arch_fc", "hallucination"] %}
- match:
prefix: "/"
headers:
- name: "x-arch-upstream"
string_match:
exact: model_server
exact: {{ internal_clustrer }}
route:
auto_host_rewrite: true
cluster: model_server
timeout: 60s
- match:
prefix: "/"
headers:
- name: "x-arch-upstream"
string_match:
exact: arch_fc
route:
auto_host_rewrite: true
cluster: model_server
cluster: {{ internal_clustrer }}
timeout: 60s
{% endfor %}
{% for _, cluster in arch_clusters.items() %}
- match:
prefix: "/"
@ -278,13 +272,14 @@ static_resources:
typed_config:
"@type": type.googleapis.com/envoy.extensions.transport_sockets.tls.v3.UpstreamTlsContext
sni: api.mistral.ai
- name: model_server
{% for internal_clustrer in ["embeddings", "zeroshot", "guard", "arch_fc", "hallucination"] %}
- name: {{ internal_clustrer }}
connect_timeout: 5s
type: STRICT_DNS
dns_lookup_family: V4_ONLY
lb_policy: ROUND_ROBIN
load_assignment:
cluster_name: model_server
cluster_name: {{ internal_clustrer }}
endpoints:
- lb_endpoints:
- endpoint:
@ -292,7 +287,8 @@ static_resources:
socket_address:
address: host.docker.internal
port_value: 51000
hostname: "model_server"
hostname: {{ internal_clustrer }}
{% endfor %}
- name: mistral_7b_instruct
connect_timeout: 5s
type: STRICT_DNS
@ -308,21 +304,6 @@ static_resources:
address: mistral_7b_instruct
port_value: 10001
hostname: "mistral_7b_instruct"
- name: arch_fc
connect_timeout: 5s
type: STRICT_DNS
dns_lookup_family: V4_ONLY
lb_policy: ROUND_ROBIN
load_assignment:
cluster_name: arch_fc
endpoints:
- lb_endpoints:
- endpoint:
address:
socket_address:
address: host.docker.internal
port_value: 51000
hostname: "arch_fc"
{% for _, cluster in arch_clusters.items() %}
- name: {{ cluster.name }}
{% if cluster.connect_timeout -%}