osd: remove redudant call of heartbeat_check

[liupan1111]

…dy called in tick_without_osd_lock.

Signed-off-by: Pan Liu pan.liu@istuary.com

[liupan1111]

@liewegas , please help take a look. thanks.

[yuyuyu101]

@liupan1111 plz remove extra rocksdb submodule updaate

[yuyuyu101]

rocksdb submodule is updated by accident

[liupan1111]

@yuyuyu101, thanks a lot for your remind. I just revert the changes for rocksdb. Please take a look.

[yuyuyu101]

@liewegas @tchaikov I give a look at another heartbeat_check, it's called under is_active and is_waiting_for_healthy. I'm concerning IS_BOOTING state also needs heartbeat_check... Although I don't know whether it's problem...

[liupan1111]

@tchaikov, any comment about this pr?

[tchaikov]

I give a look at another heartbeat_check, it's called under is_active and is_waiting_for_healthy. I'm concerning IS_BOOTING state also needs heartbeat_check... Although I don't know whether it's problem...

this heartbeat_check() call is scheduled in the tick_without_osd_lock() with the interval of osd_heartbeat_interval, with its own thread.

while the heartbeat_check() call to be removed lives in osd->heartbeat_entry(), which is in turn called in heartbeat_thread at the interval of 0.5 + random(0.. osd_heartbeat_interval), with its own thread also.

both threads are started right after the monclient starts, and before the osdmap is read from disk. so removing the heartbeat_check() call in heartbeat() won't interfere with IS_BOOTING or the boot process. however, my only concern is, prior to this change, we check the heartbeat with a random interval between 0.5 and osd_heartbeat_interval. and after this change, we check the heartbeat with a fixed interval of osd_heartbeat_interval.

but since the default osd_heartbeat_grace (20) is far larger than the default osd_heartbeat_interval (6), i think this change does not impact the behavior a lot. moreover, we have osd_fast_fail_on_connection_refused. so i think it's a safe change.

[tchaikov]

just out of curiosity, why not removing the call in tick_without_osd_lock()?

[yuriw]

http://pulpito.ceph.com/yuriw-2016-11-28_20:50:22-rados-wip-yuri-testing_2016_11_28-distro-basic-smithi/#