Fluentd 从 Docker 捕获堆栈跟踪

Question

我正在尝试让 Fluentd 解析来自 Docker 日志记录驱动程序的 Java 堆栈跟踪，使用 in_tail 并将它们作为单个消息发出。

对于我的一生，无法弄清楚为什么它仍然在分裂他们。

这是一个示例输入，正在写入文件：

2015-12-17T19:19:47+00:00   docker.java.ubuntu:15.10    {"log":"Exception in thread main java.lang.NullPointerException\r","container_id":"5a064eb23465350a11fe00b1f7787f5bd3e9f0182dd44c09516a72ab4006bd54","container_name":"/src-test_1.0.0.353_989549167.1","source":"stdout"}
2015-12-17T19:19:47+00:00   docker.java.ubuntu:15.10    {"container_id":"5a064eb23465350a11fe00b1f7787f5bd3e9f0182dd44c09516a72ab4006bd54","container_name":"/src-test_1.0.0.353_989549167.1","source":"stdout","log":"        at com.example.myproject.Book.getTitle(Book.java:16)\r"}
2015-12-17T19:19:47+00:00   docker.java.ubuntu:15.10    {"container_name":"/src-test_1.0.0.353_989549167.1","source":"stdout","log":"        at com.example.myproject.Author.getBookTitles(Author.java:25)\r","container_id":"5a064eb23465350a11fe00b1f7787f5bd3e9f0182dd44c09516a72ab4006bd54"}
2015-12-17T19:19:47+00:00   docker.java.ubuntu:15.10    {"container_id":"5a064eb23465350a11fe00b1f7787f5bd3e9f0182dd44c09516a72ab4006bd54","container_name":"/src-test_1.0.0.353_989549167.1","source":"stdout","log":"        at com.example.myproject.Bootstrap.main(Bootstrap.java:14)\r"}
2015-12-17T19:19:47+00:00   docker.java.ubuntu:15.10    {"container_id":"5a064eb23465350a11fe00b1f7787f5bd3e9f0182dd44c09516a72ab4006bd54","container_name":"/src-test_1.0.0.353_989549167.1","source":"stdout","log":"test\r"}

这是我用于 in_tail 的配置：

<source>
  @type tail
  tag docker.multiline
  path /tmp/fluent/java*
  pos_file /tmp/fluent/log.pos
  refresh_interval 10
  format multiline
  format first_line /.*\"log\":\"[^\s].*/
  format /\"log\":\"(?<message>.+)\\r/
</source>

正则表达式对我来说看起来是正确的，当我将它们插入正则表达式测试器时，first_line 正则表达式仅匹配示例的第一行和最后一行，而格式正则表达式匹配每一行，但仅捕获堆栈跟踪信息，如我期待着。但是，它们都是作为单独的消息出来的，几乎就像 first_line 匹配每一行，而不是第一行和最后一行。

Answer 1

根据https://docs.fluentd.org/v0.12/articles/parser_multiline，配置键应该是format_firstline和format（而不是format first_line和format）。