docker: do not enable eBPF by default :(

It requires Linux 5.8+ and this may trigger some support request for
little gain.

.github/e2e.hurl

@@ -67,14 +67,6 @@ inlet_sentflows: jsonpath "$.data.result[0].value[1]" toInt
 [Asserts]
 variable "inlet_sentflows" >= {{ inlet_receivedflows }}
 
-## Test inlet has eBPF running
-GET http://127.0.0.1:8080/prometheus/api/v1/query
-[Query]
-query: akvorado_inlet_flow_input_udp_ebpf_loaded{job="akvorado-inlet"}
-HTTP 200
-[Asserts]
-jsonpath "$.data.result[0].value[1]" toInt == 1
-
 ## Check outlet received flows from Kafka
 GET http://127.0.0.1:8080/prometheus/api/v1/query
 [Query]

console/data/docs/05-troubleshooting.md

@@ -431,6 +431,7 @@ In the example above, there were 486525 drops. You can solve this in different w
    is often near the value of `kafka`→`queue-size`, increase this value. This
    setting has the most impact.
 1. Increase the number of workers for the UDP input.
+1. Enable the eBPF load balancer on Linux (check `docker/docker-compose-local.yml`).
 1. Increase the value of the `net.core.rmem_max` sysctl (on the host) and
    increase the `receive-buffer` setting for the input to the same value,
 1. Increase the number of Kafka brokers.

console/data/docs/99-changelog.md

@@ -12,24 +12,13 @@ identified with a specific icon:
 
 ## Unreleased
 
-This release brings per-packet load-balancing for incoming flows using eBPF on
+If you were experiencing packet loss in the inlet after upgrading to 2.0.x,
-Linux. Previously, exporters were tied to one worker, triggering packet loss
+increase the value of `inlet`→`kafka`→`queue-size`, which was set too low. The
-despite many workers available. This requires Linux 5.8+ (support of `CAP_BPF`).
+new shipped value is 4096 instead of 32.
-For older kernels, add this to your `docker/docker-compose-local.yaml`:
-
-```yaml
-services:
-  akvorado-inlet:
-    cap_add: !override
-      - SYS_ADMIN # BPF capability is supported only from Linux 5.8
-```
-
-However, if you were experiencing packet loss in the inlet after upgrading to
-2.0.x, increase the value of `inlet`→`kafka`→`queue-size`, which was set too
-low. The new shipped value is 4096 instead of 32.
 
 - 💥 *config*: stop shipping demo exporter configurations from the orchestrator
-- 💥 *inlet*: load-balance incoming UDP packets to all workers using eBPF on Linux
+- ✨ *inlet*: load-balance incoming UDP packets to all workers using eBPF on
+  Linux (check `docker/docker-compose-local.yaml` to enable)
 - 🩹 *inlet*: fix `akvorado_inlet_flow_input_udp_in_dropped_packets_total` metric
 - 🩹 *console*: fix completion tooltip being wrapped with Firefox
 - 🌱 *config*: increase `inlet`→`kafka`→`queue-size` to a more sensible value

docker/docker-compose-local.yml

@@ -110,9 +110,9 @@
 #         - subnet: 2001:db8::/64
 #           gateway: 2001:db8::1
 
-# If you are running an older kernel than Linux 5.8, you need to uncomment the following block:
+# You can enable the eBPF load balancer to evenly spread load between workers:
 
 # services:
 #   akvorado-inlet:
-#     cap_add: !override
+#     cap_add:
-#       - SYS_ADMIN # BPF capability is supported only from Linux 5.8
+#       - BPF # or SYS_ADMIN if you don't have Linux 5.8+

docker/docker-compose.yml

@@ -155,8 +155,6 @@ services:
       kafka:
         condition: service_healthy
     command: inlet http://akvorado-orchestrator:8080
-    cap_add:
-      - BPF
     volumes:
       - akvorado-run:/run/akvorado
     labels: