There are at least two ways:

multiline plugin

Thanks to @rickerp, he suggested multiline plugin.

The multiline parser plugin parses multiline logs. This plugin is the multiline version of regexp parser.

The multiline parser parses log with formatN and format_firstline parameters. format_firstline is for detecting the start line of the multiline log. formatN, where N's range is [1..20], is the list of Regexp format for multiline log.

Unlike other parser plugins, this plugin needs special code in input plugin e.g. handle format_firstline. So, currently, in_tail plugin works with multiline but other input plugins do not work with it.

fluent-plugin-concat plugin

As per fluentd documentation, fluent-plugin-concat solves this:

Concatenate multiple lines log messages

Application log is stored into "log" field in the records. You can concatenate these logs by using fluent-plugin-concat filter before send to destinations.

<filter docker.**>@type concatkey logstream_identity_key container_idmultiline_start_regexp /^-e:2:in `\/'/multiline_end_regexp /^-e:4:in/</filter>

Original events:

2016-04-13 14:45:55 +0900 docker.28cf38e21204: {"container_id":"28cf38e212042225f5f80a56fac08f34c8f0b235e738900c4e0abcf39253a702","container_name":"/romantic_dubinsky","source":"stdout","log":"-e:2:in `/'"}2016-04-13 14:45:55 +0900 docker.28cf38e21204: {"source":"stdout","log":"-e:2:in `do_division_by_zero'","container_id":"28cf38e212042225f5f80a56fac08f34c8f0b235e738900c4e0abcf39253a702","container_name":"/romantic_dubinsky"}2016-04-13 14:45:55 +0900 docker.28cf38e21204: {"source":"stdout","log":"-e:4:in `<main>'","container_id":"28cf38e212042225f5f80a56fac08f34c8f0b235e738900c4e0abcf39253a702","container_name":"/romantic_dubinsky"}

Filtered events:

2016-04-13 14:45:55 +0900 docker.28cf38e21204: {"container_id":"28cf38e212042225f5f80a56fac08f34c8f0b235e738900c4e0abcf39253a702","container_name":"/romantic_dubinsky","source":"stdout","log":"-e:2:in `/'\n-e:2:in `do_division_by_zero'\n-e:4:in `<main>'"}

With the plugin, you'll want to fix some regexes.