DEBUG: fd: add a counter of takeovers of an FD since it was last opened

That's essentially in order to help with debugging strange cases like
the occasional epoll issues/races, by keeping a counter of how many
times an FD was taken over since last inserted. The room is available
so let's use it. If it's needed later, this patch can easily be reverted.
The counter is also reported in "show fd" as "tkov".
This commit is contained in:
Willy Tarreau 2025-01-30 15:59:11 +01:00
parent b849ee5fa3
commit 44ac7a7e73
4 changed files with 14 additions and 1 deletions

View File

@ -193,6 +193,11 @@ struct fdtab {
void *owner; /* the connection or listener associated with this fd, NULL if closed */
unsigned int state; /* FD state for read and write directions (FD_EV_*) + FD_POLL_* */
unsigned int refc_tgid; /* refcounted tgid, updated atomically */
/* the info below are mainly used for epoll debugging/strengthening.
* they're filling the rest of the cache line but may easily be dropped
* if the room is needed for more important stuff.
*/
unsigned int nb_takeover; /* number of times this FD was taken over since inserted (used for debugging) */
#ifdef DEBUG_FD
unsigned int event_count; /* number of events reported */
#endif

View File

@ -500,6 +500,10 @@ static inline void fd_insert(int fd, void *owner, void (*iocb)(int fd), int tgid
fdtab[fd].iocb = iocb;
fdtab[fd].state = newstate;
fdtab[fd].thread_mask = thread_mask;
/* just for debugging: how many times taken over since last fd_insert() */
fdtab[fd].nb_takeover = 0;
fd_drop_tgid(fd);
#ifdef DEBUG_FD

View File

@ -1415,7 +1415,7 @@ static int cli_io_handler_show_fd(struct appctx *appctx)
suspicious = 1;
chunk_printf(&trash,
" %5d : st=0x%06x(%c%c %c%c%c%c%c W:%c%c%c R:%c%c%c) ref=%#x gid=%d tmask=0x%lx umask=0x%lx prmsk=0x%lx pwmsk=0x%lx owner=%p iocb=%p(",
" %5d : st=0x%06x(%c%c %c%c%c%c%c W:%c%c%c R:%c%c%c) ref=%#x gid=%d tmask=0x%lx umask=0x%lx prmsk=0x%lx pwmsk=0x%lx owner=%p tkov=%u iocb=%p(",
fd,
fdt.state,
(fdt.state & FD_CLONED) ? 'C' : 'c',
@ -1437,6 +1437,7 @@ static int cli_io_handler_show_fd(struct appctx *appctx)
polled_mask[fd].poll_recv,
polled_mask[fd].poll_send,
fdt.owner,
fdt.nb_takeover,
fdt.iocb);
resolve_sym_name(&trash, NULL, fdt.iocb);

View File

@ -542,6 +542,9 @@ int fd_takeover(int fd, void *expected_owner)
*/
fd_stop_recv(fd);
/* essentially for debugging */
fdtab[fd].nb_takeover++;
/* we're done with it */
HA_ATOMIC_AND(&fdtab[fd].running_mask, ~ti->ltid_bit);