[
https://issues.apache.org/jira/browse/TS-3974?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
David Carlin updated TS-3974:
-----------------------------
Description:
A couple suggestions to improve the readability of the debug logs.
# Can the transaction ID be put on every debug log entry, similar to how the
thread ID is on every debug log entry? Then you could grep for a particular
transaction ID and only see those messages.
# Can we get a squid.log field for transaction ID? Then you can go looking for
it in debug log.
# Some messages currently have transaction ID inside of square braces, but
square braces are used elsewhere. For example:
\\
{noformat}[Oct 19 15:56:16.140] Server {0x2b3dbf130700} DEBUG: (http) [147596]
State Transition: SM_ACTION_ORIGIN_SERVER_OPEN ->
SM_ACTION_SERVER_READ{noformat}
In the above example 147596 is transaction ID. But look at this:
\\
{noformat}[Oct 19 10:29:10.273] Server {0x7ff0c9de08c0} DEBUG:
<HttpClientSession.cc:498 (release)> (http_cs) [2] session released by sm
[1]{noformat}
Square braces notation also used for HTTP2 client session and state machine.
So please pick something unique for transaction ID.
# Is it necessary for http header debug messages to be split across so many
lines? At a minimum can the header name and header value be on the same line?
Its hard to read debug log output when there are many clients making requests
at same time.
{noformat}
[Oct 19 20:18:44.744] Server {0x2b871b635700} DEBUG: (http)
[Oct 19 20:18:44.744] Server {0x2b871b635700} DEBUG: (http) SLOT # 4
(0x2b8818421188), LIVE
[Oct 19 20:18:44.744] Server {0x2b871b635700} DEBUG: (http) [N:
"Accept-Language", N_LEN: 15, N_IDX: 2,
[Oct 19 20:18:44.744] Server {0x2b871b635700} DEBUG: (http) V: "en-ca", V_LEN:
5,
[Oct 19 20:18:44.745] Server {0x2b871b635700} DEBUG: (http) NEXTDUP: (nil),
RAW: 1, RAWLEN: 24, F: 1]
[Oct 19 20:18:44.745] Server {0x2b871b635700} DEBUG: (http)
{noformat}
was:
A couple suggestions to improve the readability of the debug logs.
# Can the transaction ID be put on every debug log entry, similar to how the
thread ID is on every debug log entry? Then you could grep for a particular
transaction ID and only see those messages.
# Can we get a squid.log field for transaction ID? Then you can go looking for
it in debug log.
# Some messages currently have transaction ID inside of square braces, but
square braces are used elsewhere. For example:
\\
{noformat}[Oct 19 15:56:16.140] Server {0x2b3dbf130700} DEBUG: (http) [147596]
State Transition: SM_ACTION_ORIGIN_SERVER_OPEN ->
SM_ACTION_SERVER_READ{noformat}
In the above example 147596 is transaction ID. But look at this:
\\
{noformat}[Oct 19 10:29:10.273] Server {0x7ff0c9de08c0} DEBUG:
<HttpClientSession.cc:498 (release)> (http_cs) [2] session released by sm
[1]{noformat}
Square braces notation also used for HTTP2 client session and state machine.
So please pick something unique for transaction ID.
# Is it necessary for this http header debug messages to be split across so
many lines? At a minimum can the header name and header value be on the same
line? Its hard to read debug log output when there are many clients making
requests at same time.
{noformat}
[Oct 19 20:18:44.744] Server {0x2b871b635700} DEBUG: (http)
[Oct 19 20:18:44.744] Server {0x2b871b635700} DEBUG: (http) SLOT # 4
(0x2b8818421188), LIVE
[Oct 19 20:18:44.744] Server {0x2b871b635700} DEBUG: (http) [N:
"Accept-Language", N_LEN: 15, N_IDX: 2,
[Oct 19 20:18:44.744] Server {0x2b871b635700} DEBUG: (http) V: "en-ca", V_LEN:
5,
[Oct 19 20:18:44.745] Server {0x2b871b635700} DEBUG: (http) NEXTDUP: (nil),
RAW: 1, RAWLEN: 24, F: 1]
[Oct 19 20:18:44.745] Server {0x2b871b635700} DEBUG: (http)
{noformat}
> Improvements to debug log
> -------------------------
>
> Key: TS-3974
> URL: https://issues.apache.org/jira/browse/TS-3974
> Project: Traffic Server
> Issue Type: Improvement
> Components: Logging
> Reporter: David Carlin
> Labels: yahoo
>
> A couple suggestions to improve the readability of the debug logs.
> # Can the transaction ID be put on every debug log entry, similar to how the
> thread ID is on every debug log entry? Then you could grep for a particular
> transaction ID and only see those messages.
> # Can we get a squid.log field for transaction ID? Then you can go looking
> for it in debug log.
> # Some messages currently have transaction ID inside of square braces, but
> square braces are used elsewhere. For example:
> \\
> {noformat}[Oct 19 15:56:16.140] Server {0x2b3dbf130700} DEBUG: (http)
> [147596] State Transition: SM_ACTION_ORIGIN_SERVER_OPEN ->
> SM_ACTION_SERVER_READ{noformat}
> In the above example 147596 is transaction ID. But look at this:
> \\
> {noformat}[Oct 19 10:29:10.273] Server {0x7ff0c9de08c0} DEBUG:
> <HttpClientSession.cc:498 (release)> (http_cs) [2] session released by sm
> [1]{noformat}
> Square braces notation also used for HTTP2 client session and state machine.
> So please pick something unique for transaction ID.
> # Is it necessary for http header debug messages to be split across so many
> lines? At a minimum can the header name and header value be on the same
> line? Its hard to read debug log output when there are many clients making
> requests at same time.
> {noformat}
> [Oct 19 20:18:44.744] Server {0x2b871b635700} DEBUG: (http)
> [Oct 19 20:18:44.744] Server {0x2b871b635700} DEBUG: (http) SLOT # 4
> (0x2b8818421188), LIVE
> [Oct 19 20:18:44.744] Server {0x2b871b635700} DEBUG: (http) [N:
> "Accept-Language", N_LEN: 15, N_IDX: 2,
> [Oct 19 20:18:44.744] Server {0x2b871b635700} DEBUG: (http) V: "en-ca",
> V_LEN: 5,
> [Oct 19 20:18:44.745] Server {0x2b871b635700} DEBUG: (http) NEXTDUP: (nil),
> RAW: 1, RAWLEN: 24, F: 1]
> [Oct 19 20:18:44.745] Server {0x2b871b635700} DEBUG: (http)
> {noformat}
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
