fix(eks-addons): let Loki return query results larger than 4MiB

README.md

@@ -82,7 +82,7 @@ sync by `node tools/ravion-modules/dist/src/cli.js readme` (enforced in CI, and
 | `rvn-ecs-web` | ECS Web Service | v2.3.0 | `compute/ecs_service/` |
 | `rvn-ecs-worker` | ECS Worker | v2.2.2 | `compute/ecs_service/` |
 | `rvn-efs` | EFS File System | v1.1.0 | `storage/efs/` |
-| `rvn-eks-addons` | EKS Add-ons | v0.23.0 | `compute/eks/addons/` |
+| `rvn-eks-addons` | EKS Add-ons | v0.23.2 | `compute/eks/addons/` |
 | `rvn-eks-chart` | EKS Helm Chart | v0.3.0 | `compute/eks_service/` |
 | `rvn-eks-cluster` | EKS Cluster | v0.9.0 | `compute/eks/` |
 | `rvn-eks-web` | EKS Web Service | v1.10.0 | `compute/eks_service/` |

compute/eks/addons/loki.tf

@@ -218,6 +218,14 @@ resource "helm_release" "loki" {
 
           server = {
             http_listen_port = 3100
+
+            # In single-binary mode the querier returns each result to the
+            # query frontend over gRPC, and a filter query over a busy
+            # workload's logs easily passes the 4MiB default, failing with
+            # "received message larger than max". 100MiB matches the querier's
+            # own client limit.
+            grpc_server_max_recv_msg_size = 104857600
+            grpc_server_max_send_msg_size = 104857600
           }
 
           # A schema is the one thing the chart refuses to guess, because

compute/eks/addons/rvn-eks-addons-definition.yml

@@ -4,8 +4,8 @@ definition:
   name: EKS Add-ons
   description: Selectable add-ons for an existing EKS cluster.
 release:
-  version: 0.23.0
-  description: "Mimir ingesters drop series from pods that have gone, compacting early once they hold half their memory in series, so rollouts and scale-downs no longer fill them until they are killed for memory. Ingesters and store-gateways now refuse to share a node, as the three copies promise."
+  version: 0.23.2
+  description: "Loki answers log searches whose results pass 4MiB. Its gRPC server kept the 4MiB message default, so a filter over a busy workload's logs failed with \"received message larger than max\"; it now accepts up to 100MiB, the querier's own limit."
 module:
   inputs:
     - id: cluster

compute/eks/addons/tests/logs.tftest.hcl

@@ -298,6 +298,17 @@ run "loki_runs_as_a_single_binary_with_no_extras" {
     error_message = "The gateway, both memcached tiers, and the bundled MinIO must all be off"
   }
 
+  # In single-binary mode the querier hands each result back to the query
+  # frontend over gRPC. At the 4MiB default, a filter query over a busy
+  # workload's logs fails with "received message larger than max".
+  assert {
+    condition = alltrue([
+      yamldecode(helm_release.loki[0].values[0]).loki.server.grpc_server_max_recv_msg_size == 104857600,
+      yamldecode(helm_release.loki[0].values[0]).loki.server.grpc_server_max_send_msg_size == 104857600,
+    ])
+    error_message = "Loki's gRPC server must accept query results larger than the 4MiB default."
+  }
+
   # Structured metadata is where `level` lives; without this the schema rejects
   # it and the whole label design collapses into index labels.
   assert {