Attempt to fix some more test issues.

- Fix Fluent Bit HTTP proxy test.
- Attempt to fix differing behavior of invalid utf-8 handling in CI.

templates/etc/fluent-bit/fluent-bit.yaml.etlua

@@ -11,6 +11,8 @@ service:
   # Determine how many MB of data can be read from disk in the event fluent-bit
   # is restarted.
   storage.backlog.mem_limit: <%- json_encode(config["fluent_bit"]["service"]["storage_backlog_mem_limit"]) %>
+  # Enable hot reloading for some testing scenarios.
+  hot_reload: on
 
 pipeline:
   inputs:

templates/etc/perp/fluent-bit/rc.env.etlua

@@ -1,6 +1,3 @@
 <% if config["http_proxy"] then %>
-http_proxy=<%- config["http_proxy"] %>
-<% end %>
-<% if config["https_proxy"] then %>
-https_proxy=<%- config["https_proxy"] %>
+HTTP_PROXY=<%- config["http_proxy"] %>
 <% end %>

test/proxy/envoy/test_http_proxy.rb

@@ -92,16 +92,23 @@ def test_opensearch_requests_use_proxy
       record = wait_for_log(response)[:hit_source]
       assert_equal("/api/hello", record["request_path"])
 
-      log_output = log_tail.read_until(%r{"uri":"/_bulk"})
-      log = MultiJson.load(log_output.scan(%r{^.*"uri":"/_bulk".*$}).last)
-      assert_equal("/_bulk", log.fetch("uri"))
+      # Fluent Bit establishes a CONNECT tunnel for its proxying, so the only
+      # thing being logged in the HTTP access log is the initial CONNECT
+      # request. So we won't actually see the /_bulk requests being sent over
+      # this tunnel. Also send a SIGHUP to fluent-bit, since otherwise it seems
+      # like it can take a while for this CONNECT request to be flushed to the
+      # logs (probably since it's a persistent tunnel).
+      api_umbrella_process.perp_signal("fluent-bit", "hup")
+      log_output = log_tail.read_until(%r{"user_agent":"Fluent-Bit"})
+      log = MultiJson.load(log_output.scan(%r{^.*"user_agent":"Fluent-Bit".*$}).last)
+      assert_nil(log["uri"])
       assert_equal("opensearch:9200", log.fetch("host"))
       assert_equal("http", log.fetch("scheme"))
-      assert_equal("POST", log.fetch("method"))
+      assert_equal("CONNECT", log.fetch("method"))
       assert_equal(200, log.fetch("status"))
       assert_match(/\A[0-9a-f.:]+:9200\z/, log.fetch("up_addr"))
       assert_nil(log["up_tls_ver"])
-      assert_nil(log["user_agent"])
+      assert_equal("Fluent-Bit", log.fetch("user_agent"))
 
       response = Typhoeus.get("https://127.0.0.1:9081/admin/stats/logs.json", http_options.deep_merge(admin_session).deep_merge({
         :params => {

test/proxy/logging/test_special_chars.rb

@@ -128,14 +128,32 @@ def test_invalid_utf8_encoding_in_url_path_url_params_headers
     record = wait_for_log(response)[:hit_source]
 
     log = log_tail.read.encode("UTF-8", invalid: :replace)
-    assert_match("invalid UTF-8 bytes found, skipping bytes", log)
+    # Fluent Bit's UTF-8 handling appears to be different on x86_64 versus
+    # ARM64. I believe related to this open issue:
+    # https://github.com/fluent/fluent-bit/issues/7995
+    #
+    # So on ARM64, it complains about UTF-8, which I think is more expected
+    # since we are sending in invalid encoded data. But on x86-64 systems, this
+    # doesn't appear to happen. I think we're okay with either behavior,
+    # really, we mainly just want to make sure things don't crash when
+    # encountering this type of weird input.
+    if RUBY_PLATFORM.start_with?("x86_64")
+      refute_match("invalid UTF-8 bytes found, skipping bytes", log)
+    else
+      assert_match("invalid UTF-8 bytes found, skipping bytes", log)
+    end
 
     # Since the encoding of this string wasn't actually a valid UTF-8 string,
     # we test situations where it's sent as the raw ISO-8859-1 value, as well
     # as the UTF-8 replacement character.
     expected_raw_in_url_path = url_encoded.downcase
     expected_raw_in_url_query = url_encoded
-    expected_raw_in_header = ""
+    # See above for differences in platform.
+    if RUBY_PLATFORM.start_with?("x86_64")
+      expected_raw_in_header = "£"
+    else
+      expected_raw_in_header = ""
+    end
     expected_raw_utf8_in_url_path = "%ef%bf%bd"
     expected_raw_utf8_in_url_query = "%EF%BF%BD"
     expected_raw_utf8_in_header = Base64.decode64("77+9").force_encoding("utf-8")