[LU-16800] Improve Lustre API error reporting

• Type: Improvement
• Resolution: Unresolved
• Priority: Minor
• Severity: 3
• Assignee: Alexandre Ioffe
• Reporter: Alexandre Ioffe

Description

• Add rate control of info and error messages
• Add ability to print file name/function name/line number of where from the message is sourced from
• Eliminate the split of one message on a number printfs(). Such split may cause one error message be separated on multiple distanced lines in the log in multithread application
• Fix this code:
  lustre\utils\liblustreapi.c : error_callback_default()

	/*
		 * Remove trailing linefeed so error string can be appended.
		 * @fmt is a const string, so we can't modify it directly.
		 */
		if (has_nl && (newfmt = strdup(fmt)))
			*strrchr(newfmt, '\n') = '\0';

This code trims end of line if the format has more than one '\n'

 

We can use level parameter to pass file/function/line format specifiers.

Need to review all cases of level parameter use. Review this code:

lustre\utils\liblustreapi_hsm.c :  llapi_hsm_log_error()

 

    real_level = level & LLAPI_MSG_NO_ERRNO;
    real_level = real_level > 0 ? level - LLAPI_MSG_NO_ERRNO : level;
 

 

Comments

[Gerrit Updater]

"Alexandre Ioffe <[e-mail]>" uploaded a new patch: https://review.whamcloud.com/c/fs/lustre-release/+/56175
Subject: LU-16800 lustre: Improve Lustre API error reporting
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 1aa08988c576c7baae454d057b7689db51703736

[Alexandre Ioffe]

Review in 6.0

https://review.whamcloud.com/c/ex/lustre-release/+/50859/

 

[Alexandre Ioffe]

To test and demonstrate different aspects of the new error logging API I have integrated it in lamigo/lpurge. This example does not have the rate control.

Here are samples (subject to further correction as needed): 

Used macro

 

LLAPI_PRINTF((LLAPI_MSG_DEBUG | LX_PRINTF_OPT), lx_log_prefix, 0, fmt, ##args)

and

 

 

LLAPI_PRINTF((LLAPI_MSG_INFO | LX_PRINTF_OPT),  lx_log_prefix, 0, fmt, ##args)

 

 

LX_PRINTF_OPT =  LLAPI_MSG_NO_ERRNO | LLAPI_MSG_PREFIX | LLAPI_MSG_SEVERITY | LLAPI_MSG_TIMEOFDAY | LLAPI_MSG_FILE | LLAPI_MSG_LINE

 

 

lx_log_prefix - custom prefix "testfs-MDT0000"

 

1685086614.273859 testfs-MDT0000:DEBUG:lamigo.c:3527:sync hot to fast [0x200000401:0x3ff1:0x0]: H: 0/1, P: 1/0, L 1, I 0
1685086594.896799 testfs-MDT0000:DEBUG:lamigo_alr.c:192:keepalive msg from host:'ost-centOS8'
1685086622.056213 testfs-MDT0000:INFO:lamigo.c:3355:received signal 15, exiting

The same as above, but no LLAPI_MSG_FILE | LLAPI_MSG_LINE

 

1685127033.955474 testfs-MDT0000:DEBUG:sync hot to fast [0x200000401:0x5:0x0]: H: 0/1, P: 0/1, L 1, I 0

 

 

Backward compatible calls (prepended by llapi_set_command_name(opt.o_mdtname). The output is supposedly the same as original:

 

llapi_error(LLAPI_MSG_INFO, 0, fmt, ##args);

 

 

lamigo testfs-MDT0000: sync hot to fast [0x200000401:0x5:0x0]: H: 0/1, P: 0/1, L 1, I 0 : Success (0)

 

 

 

llapi_err_noerrno(LLAPI_MSG_INFO, fmt, ##args);

 

 

lamigo testfs-MDT0000: sync hot to fast [0x200000401:0x5:0x0]: H: 0/1, P: 0/1, L 1, I 0

 

 

 

llapi_printf(LLAPI_MSG_INFO, fmt, ##args);

 

 

sync hot to fast [0x200000401:0x5:0x0]: H: 0/1, P: 0/1, L 1, I 0