mon/MonClient: respect priority in SRV

before this change, we always choose the monitors with the lowest
priority and ignore the ones with higher priorty. but per
https://www.ietf.org/rfc/rfc2782.txt

> A client MUST attempt to contact the target host with the
> lowest-numbered priority it can reach

so, if a monitor is not reachable, we should try the ones with the
lowest ones *reachable*.

after this change, MonClient will memorize the monitors have been
contacted. and will continue hunting the ones not tried if the last
batch fail.

Signed-off-by: Kefu Chai <kchai@redhat.com>

Merge pull request #25545 from zhouyuan/wip-ceph-immutable-obj-cache-daemon

tools: adding ceph level immutable obj cache daemon

Reviewed-by: Jason Dillaman <dillaman@redhat.com>
Reviewed-by: Mykola Golub <mgolub@suse.com>

Merge pull request #26675 from trociny/wip-rbd-io-simple-scheduler

librbd: simple scheduler plugin for object dispatcher layer

Reviewed-by: Jason Dillaman <dillaman@redhat.com>

Merge pull request #27098 from ming416/master

mgr/dashboard: auth ttl expired error

Reviewed-by: Ricardo Dias <rdias@suse.com>
Reviewed-by: Volker Theile <vtheile@suse.com>

Merge PR #27070 into master

* refs/pull/27070/head:
osd/OSDMap: add 'zone' to default crush map

Reviewed-by: Greg Farnum <gfarnum@redhat.com>

Merge PR #27020 into master

* refs/pull/27020/head:
common/blkdev: handle devices with ID_MODEL as "LVM PV ..." but valid ID_MODEL_ENC

Reviewed-by: Kefu Chai <kchai@redhat.com>

Merge PR #27021 into master

* refs/pull/27021/head:
msg: remove XioMessenger
qa/suites/rados/thrash-old-clients: add nautilus
qa/suites/rados/thrash-old-clients: add mimic v1 variant
qa/suites/rados/thrash-old-clients: add mimic
qa/suites/rados/thrash-old-clients: collapse msgr and client choice
qa: remove simplemessenger tests
ceph_test_msgr: remove simple
msg: remove SimpleMessenger

Reviewed-by: xie xingguo <xie.xingguo@zte.com.cn>
Reviewed-by: Matt Benjamin <mbenjami@redhat.com>
Reviewed-by: Kefu Chai <kchai@redhat.com>

Merge PR #27101 into master

* refs/pull/27101/head:
mon/MgrStatMonitor: ensure only one copy of initial service map

Reviewed-by: Neha Ojha <nojha@redhat.com>

mgr/dashboard: auth ttl expired error

Fixes: https://tracker.ceph.com/issues/38428
Signed-off-by: ming416 <geenature@163.com>

Merge pull request #26940 from xiexingguo/wip-monc-add-con

mon/MonClient: weight-based mon selection

Reviewed-by: Kefu Chai <kchai@redhat.com>

Merge pull request #26955 from liewegas/wip-slow-add

crush: various fixes for weight-sets, the osd_crush_update_weight_set option, and tests

Reviewed-by: xie xingguo <xie.xingguo@zte.com.cn>

Merge pull request #26950 from xiexingguo/wip-mgr-fixes

mgr/Mgr: kill redundant sub_unwant call

Reviewed-by: Kefu Chai <kchai@redhat.com>

Merge pull request #27068 from hjwsm1989/upmap-filter-out-underfull-buckets

crush: add root_bucket to identify underfull buckets

Reviewed-by: Sage Weil <sage@redhat.com>
Reviewed-by: xie xingguo <xie.xingguo@zte.com.cn>

Merge pull request #27067 from SUSE/wip-fix-38788

cmake: remove cython 0.29's subinterpreter check during install

Reviewed-by: Kefu Chai <kchai@redhat.com>

Merge pull request #26923 from majianpeng/msg-optimize-check-loopback-con

msg/async/AsyncConnection: optimize check loopback connection.

Reviewed-by: Ricardo Dias <rdias@suse.com>

Merge pull request #27076 from MariusSchiffer/ceph-mgr-python3-fix

mgr/balancer: Python 3 compatibility fix

Reviewed-by: Kai Wagner <kwagner@suse.com>
Reviewed-by: Kefu Chai <kchai@redhat.com>

Merge pull request #27106 from dalgaaf/dalgaaf-wip-fix-lrc-doc

doc: fix LRC documentation

Reviewed-by: Kefu Chai <kchai@redhat.com>

Merge pull request #27071 from tchaikov/wip-crimson-osd-peering

crimson/osd: add minimal state machine for PG peering

Reviewed-by: Radoslaw Zarzynski <rzarzyns@redhat.com>
Reviewed-by: Samuel Just <sjust@redhat.com>

crimson/osd: add minimal PG recovery FSM

after this change, PG is able to go clean as primary

Signed-off-by: Kefu Chai <kchai@redhat.com>

crimson/osd: advance pg in consume_map()

Signed-off-by: Kefu Chai <kchai@redhat.com>

crimson/osd/pg: update info.stats

PG.info.stats is sent to mgr periodically for PG stats reporting,
also, it is sent to peer OSD in MOSDPGLog message when PG performs
peering. we need to keep it updated.

Signed-off-by: Kefu Chai <kchai@redhat.com>

crimson/osd: add more peering facilities

* add PG::should_restart_peering() for telling if we should start
  a peering at seeing a new mapping
* add PG::start_peering_interval() for starting a peering
* add PG::activate() for activating a PG
* add PG::on_activated() to be called once a PG is activated
* add PG::send_to_osd() for sending message to given OSD

Signed-off-by: Kefu Chai <kchai@redhat.com>

crimson/osd: add operator<< for PG

Signed-off-by: Kefu Chai <kchai@redhat.com>

crimson/osd/pg: add peer_activated

add facilities for updating peer_activated. it is used for tracking the
peers has activated and committed. once all of them ack the activation
proposed by primary, the PG is considered activated.

Signed-off-by: Kefu Chai <kchai@redhat.com>

crimson/osd: add facilities for sending notify

we need to notify the primary osd that a PG hosted by it has changed,
and it needs to start peering for it.

Signed-off-by: Kefu Chai <kchai@redhat.com>

crimson/osd: maintain PG::want_acting

* add PG::choose_acting()
* add PG::proc_replica_info()

Signed-off-by: Kefu Chai <kchai@redhat.com>

crimson/osd: add PG::is_{acting,up}()

Signed-off-by: Kefu Chai <kchai@redhat.com>

crimson/osd: wait osdmap before processing peering evt

the constraint that

> peering messages/events from peers must remain ordered by peer

is still enforced. because, the peering message from a certain peer
will block the succeeding messages from this peer, if the latest osdmap
possessed by OSD is not new enough.

Signed-off-by: Kefu Chai <kchai@redhat.com>

crimson/osd: add OSD::_send_active()

so PG can request monitor to update the OSD's up_thru in osdmap

Signed-off-by: Kefu Chai <kchai@redhat.com>

crimson/osd: handle MOSDPGLog in OSD

will handle it in PG in follow-up changes.

Signed-off-by: Kefu Chai <kchai@redhat.com>

crimson/osd: handle MOSDPGQuery in OSD

will handle it in PG in follow-up changes.

Signed-off-by: Kefu Chai <kchai@redhat.com>

crimson/osd: handle MOSDPGInfo in OSD

will handle it in PG in follow-up changes.

Signed-off-by: Kefu Chai <kchai@redhat.com>

crimson/osd: handle MOSDPGNotify in OSD

will handle it in PG in follow-up changes.

Signed-off-by: Kefu Chai <kchai@redhat.com>

crimson/osd: keep track of last_peering_reset

last_peering_reset is used to track the epoch when the most recent peering
started. it is used to tell if a message is targeting a changed PG. the
sender of a peering message will attach it. and the recipient will check
it against its own last_peering_reset. for instance, if a message claims
that that its map epoch is 16, while our last_peering_reset is 18, we will
consider this message stale, and probably drop it.

Signed-off-by: Kefu Chai <kchai@redhat.com>

doc: fix LRC documentation

Recovery from a failure in jerasure need only k reads and
not k+m-1.

Signed-off-by: Danny Al-Gaaf <danny.al-gaaf@bisect.de>

common/blkdev: handle devices with ID_MODEL as "LVM PV ..." but valid ID_MODEL_ENC

Some devices appear like so:

E: ID_MODEL=LVM PV LUClYG-Oyte-jcM6-npfZ-ncsl-ycL0-bkOH0m on /dev/sdn
E: ID_MODEL_ENC=WDC\x20WDS200T2B0A-00SM50\x20\x20\x20\x20\x20\x20\x20\x20\x20\x20\x20\x20\x20\x20\x20\x20\x20\x20
E: ID_SERIAL=WDC_WDS200T2B0A-00SM50_183503800168
E: ID_SERIAL_SHORT=183503800168

To avoid upsetting our prioritization, defer to ID_MODEL_ENC over ID_MODEL
if ID_MODEL has "LVM PV" in it (as opposed to just using ID_SERIAL).

Signed-off-by: Sage Weil <sage@redhat.com>

Merge PR #27044 into master

* refs/pull/27044/head:
common: add bool log_to_file option

Reviewed-by: Kefu Chai <kchai@redhat.com>
Reviewed-by: Neha Ojha <nojha@redhat.com>

osd/OSDMap: add 'zone' to default crush map

This lets use have a root -> region -> zone -> host hierarchy, like you
will see in the public cloud.

Signed-off-by: Sage Weil <sage@redhat.com>

mon/MgrStatMonitor: ensure only one copy of initial service map

It's possible for create_pending to be called multiple times.

Fixes: http://tracker.ceph.com/issues/38839
Signed-off-by: Sage Weil <sage@redhat.com>

Merge PR #27048 into master

* refs/pull/27048/head:
doc: add changelog for nautilus

Reviewed-by: Sage Weil <sage@redhat.com>

test/librbd: sequential write aio test

Signed-off-by: Mykola Golub <mgolub@suse.com>

tools: move options parsing into ObjectCacheStore for immutable obj cache daemon

This makes the options parsing more easy to read. SimplePolicy will get those
options from ObjectCacheStore.

Signed-off-by: Yuan Zhou <yuan.zhou@intel.com>

building: adding missing ceph-immtable-object-cache-dbg rule

Signed-off-by: Yuan Zhou <yuan.zhou@intel.com>

tools: fix SimplyPolicy memory leak

Signed-off-by: Dehao Shang <dehao.shang@intel.com>

tools: cleanup code style and delete useless data

Signed-off-by: Dehao Shang <dehao.shang@intel.com>

tools: new cache sub-dir rule for immutable obj cache daemon

do not use static sub-dirs for cache, instead we create the sub-dir
dynamically with the first two charactor of crc32(file_name)

Signed-off-by: Yuan Zhou <yuan.zhou@intel.com>

tools: cleanup RO

- replace useless callback with bind function.
- remove session_id and m_session_map.
- fixed memory leak when session occur error
- change type conversion.
- add prefix

Signed-off-by: Dehao Shang <dehao.shang@intel.com>

tools: eliminate session_map race between CacheSession

Signed-off-by: Dehao Shang <dehao.shang@intel.com>

tools: make cache file dir number configurable

Signed-off-by: Dehao Shang <dehao.shang@intel.com>

tools: make cache watermark configurable

Signed-off-by: Dehao Shang <dehao.shang@intel.com>

tools: make dedicated thread configurable for RO

At cache client side of RO, make dedicated woker thread configurable
Also, cleanup useless member of CacheClient class

Signed-off-by: Dehao Shang <dehao.shang@intel.com>

tools: adjust code style of RO

Signed-off-by: Dehao Shang <dehao.shang@intel.com>

tools: clean up data structure of immutable obj cache daemon

Signed-off-by: Yuan Zhou <yuan.zhou@intel.com>

tools: refactor ObjectCacheRequest of RO

Signed-off-by: Dehao Shang <dehao.shang@intel.com>

tools: return real cache status on lookup in immutable obj cache

Signed-off-by: Yuan Zhou <yuan.zhou@intel.com>

tools: fix librados compatibility

Signed-off-by: Yuan Zhou <yuan.zhou@intel.com>

tools: clean up cache dir for immutable obj cache

- user defined full path for cache dir
- hash objects into subfolders with crc32 of its name

Signed-off-by: Yuan Zhou <yuan.zhou@intel.com>

tools: maintain cache size counter for immutable obj cache daemon

Do not assume 4M object size since we allow to cache any-sized objects.

Signed-off-by: Yuan Zhou <yuan.zhou@intel.com>

tools: templatize create_rados_callback

Also cleanup unused template codes

Signed-off-by: Dehao Shang <dehao.shang@intel.com>

tools: use specific message for different ops in immutable obj cache daemon

use different types of message for different ops to be more efficient

Signed-off-by: Yuan Zhou <yuan.zhou@intel.com>

tools: cleanup IPC message for immutable obj cache daemon

using ceph serialization(encode/decode) framework instead of building
new data structure

Signed-off-by: Yuan Zhou <yuan.zhou@intel.com>

tools: set namespace & snap_id before promote from RADOS in immutable obj cache

Signed-off-by: Yuan Zhou <yuan.zhou@intel.com>

tools: format code and clean up

Signed-off-by: Yuan Zhou <yuan.zhou@intel.com>

tools: adding signal handling for immutable obj cache daemon

Signed-off-by: Yuan Zhou <yuan.zhou@intel.com>

tests: reduce testing resources occupied for immutable obj cache

Signed-off-by: Yuan Zhou <yuan.zhou@intel.com>

tools: follow current codes to modify UT

Signed-off-by: Dehao Shang <dehao.shang@intel.com>

tools: release dedicated thread resource

Addtional, enable all UT

Signed-off-by: Dehao Shang <dehao.shang@intel.com>

tools: add connect fails to fault

Signed-off-by: Dehao Shang <dehao.shang@intel.com>

tools: clean up object cache request struture

adding version and compat_version info for serailization

Signed-off-by: Yuan Zhou <yuan.zhou@intel.com>

common/buffer: assign error message to caller on short read in pread_file

Signed-off-by: Yuan Zhou <yuan.zhou@intel.com>

tools: add m_cache_path

1) : enable ObjectCacheRequest.m_data.m_cache_path
2) : add generate_cache_file_path to create cache_path
    , which will be used by CacheClient.

Signed-off-by: Dehao Shang <dehao.shang@intel.com>

tools: clean up dead code in immutable obj cache

Signed-off-by: Yuan Zhou <yuan.zhou@intel.com>

tools: kill ObjectCacheFile in immutable obj cache

instead we could use bufferlist::write_file/pread_file to write/read on cache

Signed-off-by: Yuan Zhou <yuan.zhou@intel.com>

tools: clean up immutable obj cache

Signed-off-by: Yuan Zhou <yuan.zhou@intel.com>

common/options: move immutable obj cache options out of rbd section

Signed-off-by: Yuan Zhou <yuan.zhou@intel.com>

common/buffer: return 0 on short read in pread_file

Signed-off-by: Yuan Zhou <yuan.zhou@intel.com>

tools: some modification at RO

1) : add new data member to ObjectCacheRequest,
     including m_pool_namepace, m_pool_id, m_snap_id,
     m_cache_path
2) : add some log at CacheSession
3) : change pool name to pool_nspace + poo_id + snap_id
4) : change cache file name
5) : add generate_cache_file_name method
6) : modify UT interface

Signed-off-by: Dehao Shang <dehao.shang@intel.com>

tool: break down receive_message implements

Also, enable fault at CacheClient side.

Signed-off-by: Dehao Shang <dehao.shang@intel.com>

common/buffer: fix pread_file to generate local bufferlist

Signed-off-by: Yuan Zhou <yuan.zhou@intel.com>

tools: clean up types for immutable obj cache daemon

Signed-off-by: Yuan Zhou <yuan.zhou@intel.com>

tools: refuse to start if obj cache daemon fail to initialize cache dir

Signed-off-by: Yuan Zhou <yuan.zhou@intel.com>

test: enable asio tests for immutable obj cache controller

cleanups for immutable obj cache controller

Signed-off-by: Yuan Zhou <yuan.zhou@intel.com>

tools: immutable_object_cache code format cleanup

Signed-off-by: Yuan Zhou <yuan.zhou@intel.com>

common/buffer: adding pread_file method

pread_file allows to read length of contents at a given offset

using pread_file instead of read_file to reduce extra reads in ObjectCacheFile

Signed-off-by: Yuan Zhou <yuan.zhou@intel.com>

tools: refactor asio domain socket of RO

Signed-off-by: Dehao Shang <dehao.shang@intel.com>
Signed-off-by: Yuan Zhou <yuan.zhou@intel.com>

tools: fix redudent updates on cache metdata

Signed-off-by: Yuan Zhou <yuan.zhou@intel.com>

build/ops: adding build spec for immutable object cache daemon

Signed-off-by: Yuan Zhou <yuan.zhou@intel.com>

test: add object cache message unite test

Signed-off-by: Dehao Shang <dehao.shang@intel.com>

tools: add new object cache message

Signed-off-by: Dehao Shang <dehao.shang@intel.com>

tools: use bufferlist to implement ObjectCacheFile

1) : modify unite testing of object cache file
2) : remove open and create internface
3) : use bufferlist interface to read/write file

Signed-off-by: Dehao Shang <dehao.shang@intel.com>

test: add ObjectCacheStore unit test

Signed-off-by: Dehao Shang <dehao.shang@intel.com>

test: add multi-session unit testing

Signed-off-by: Dehao Shang <dehao.shang@intel.com>

test: move WaitEvent class to test_common.h file

Signed-off-by: Dehao Shang <dehao.shang@intel.com>

tools: fix immutable obj cache max size

Signed-off-by: Yuan Zhou <yuan.zhou@intel.com>

tools: remove RBD specific logic in immutable obj cache

Make the daemon more generic(pool name + object id) so it
can service both RBD and RGW.
Also supports caching partial object(even empty), the caller
needs to append_zero to read buffer if read on those objects.

Signed-off-by: Yuan Zhou <yuan.zhou@intel.com>

tools: remove background eviction

do eviction check in promotion callback simply

Signed-off-by: Yuan Zhou <yuan.zhou@intel.com>

tools: format headers for object cache daemon

Signed-off-by: Yuan Zhou <yuan.zhou@intel.com>

tools: create local utils for object cache daemon

Signed-off-by: Yuan Zhou <yuan.zhou@intel.com>

tools: clean up immutable object cache

- drop "ceph_" prefix
- prefix all options with immutable_object_cache_
- adding configurable immutable_object_cache_max_inflight_ops

Signed-off-by: Yuan Zhou <yuan.zhou@intel.com>

tools: adding ceph level immutable obj cache daemon

The daemon is built for future integration with both RBD and RGW cache.
The key components are:
- domain socket based simple IPC
- simple LRU policy based promotion/demotion for the cache
- simple file based caching store for RADOS objs with sync IO interface
- systemd service/target files for the daemon

Signed-off-by: Dehao Shang <dehao.shang@intel.com>
Signed-off-by: Yuan Zhou <yuan.zhou@intel.com>