rgw: a few bug fixes related to async operations

fix uninitialized state, trim stacks, etc.

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>

rgw: add a generic op to pull data off metadata obj

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>

rgw: add another state for simple async ops

so that these ops can chain other ops and handle their
response

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>

rgw: more simplification

create a helper method in the ops manager to run a single op, it
builds the needed env

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>

rgw: more simplification, no need to take reference to op

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>

rgw: some cleanup, keep async env structure

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>

rgw: create simple async op handler

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>

rgw: async rados requests through a work queue

Tie in the async rados requests to the async framework.

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>

rgw: retrieve a specific sync shard info

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>

rgw-admin: command to show mdlog sync status

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>

rgw: stackable ops infrastructure

ops can now call other ops, so that we could simplify
the different states each ops need to keep.

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>

rgw: initial work on sync status marker

Also, rename radosgw-admin mdlog sync to radosgw-admin mdlog-fetch.

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>

rgw: read mdlog marker

Use our local marker as a starting point for fetching the mater mdlog.

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>

rgw: clean up mdlog clone states, other fixes

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>

rgw: don't reference msg after finish_request()

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>

rgw: hide the completion manager

some more layer separation

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>

rgw: create some abstraction around async ops infrastructure

This will be reused later for other operations.

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>

rgw: mdlog sync error reporting on failed request

Basic error reporting. We'll need to dump this to the backend and
add admin commands to read.

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>

some error handling work, not ready yet

rgw: md log cloning completely async

hook librados completions into the rgw completion notifier, split the
md log write into two different states.

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>

rgw: rework metadata log sync to use async capabilities

We now have a window of operations. The next step will be to
asynchronously write to the local mdlog, so that we don't
just hold the work on a synchronous request.

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>

rgw: can now wait on async rest requests

In addition to waiting on a specific request, added a
completion manager that we can wait on so that we get
which request is ready.

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>

rgw: add new class to create resource request

RGWRESTReadResource holds the request input and output. Previously
we were using stack variables that couldn't be used in the async
path.

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>

rgw: more http aio stuff

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>

rgw: threaded http manager groundwork

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>

rgw: more work on cleaning up http client interfaces

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>

rgw: rework http client to handle multiple async requests

Currently we use the curl multi handler using only a single handle. Rework
things a bit to prepare for multiple handlers.

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>

rgw: paginated mdlog clone

still need to iterate by going through over older entries first

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>

rgw: initial mdlog clone implementation

naive handling, not paging, unbalanced.

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>

rgw: can list mdlog entries

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>

rgw: move json helper into RGWRESTConn

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>

ceph_json: make sure header doesn't clobber _ASSERT_H

_ASSERT_H is needed for dout() compilation

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>

rgw: decode number of mdlog shards

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>

rgw: fix use of zone name

zone_name doesn't really hold what we need

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>

rgw-admin: initial mdlog sync work

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>

rgw: initial work on mdlog sync

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>

rgw: don't do set_attr on meta objects

only complete object writes (and attributes) is now available. This helps
managing the meta heap and simplifies the states. Since we cannot use rm_attrs
anymore, need to reset meta object before writing it.

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>

rgw: store immutable metadata entries in heap pool

Define a new pool for immutable metadata, store new meta objects
in that pool.
(still need to work on the set_attrs case)

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>

rgw: rename default pool names

Name default pools a bit better

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>

rgw: don't require system pools to start with a period

Make sure cache operations are being done on system objects.
Separate calls where needed to make sure that separate pool/bucket
handling is done correctly. When separating calls, don't use objv_tracker
for data objects, don't use data constructs for system objects.

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>

cmake: add cls_timeindex to rgw targets

Signed-off-by: Casey Bodley <cbodley@redhat.com>

Merge pull request #7496 from xiexingguo/xxg-wip-fixnvme

osd: BlueStore/NVMEDevice: fix compiling and fd leak

Reviewed-by: Haomai Wang <haomai@xsky.com>

Merge pull request #7349 from yuyuyu101/add-pwritev

osd: FileStore: use pwritev instead of lseek+writev

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

Merge pull request #7504 from renhwsky/renhw-wip-mon-shutdown

mon: unregister command on shutdown

Reviewed-by: Joao Eduardo Luis <joao@suse.de>

Merge pull request #7550 from tchaikov/wip-osd-cleanup

osd: a fix for HeartbeatDispatcher and cleanups

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

Merge pull request #7553 from XinzeChi/wip-osdmap-affinity

osd: OSDMap: reset osd_primary_affinity shared_ptr when deepish_copy_from

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

Merge pull request #7560 from tchaikov/wip-fix-icl

crush/CrushTester: workaround a bug in boost::icl

Merge pull request #7568 from liewegas/wip-bluestore-mutex

osd: bluestore/blockdevice: use std::mutex et al

Merge pull request #7578 from ceph/fix-stripper

Stripper: Fix incorrect push_front -> append_zero change

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

Stripper: Fix incorrect push_front -> append_zero change

Signed-off-by: Haomai Wang <haomai@xsky.com>

Merge pull request #7444 from dreamhost/wip-rgw-static-website-robbat2

rgw: approximate AmazonS3 HostId error field.

Reviewed-by: Yehuda Sadeh <yehuda@redhat.com>

Merge remote-tracking branch 'gh/jewel'

v10.0.3

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

Merge pull request #7572 from ceph/wip-cmake-2-8-16

cmake: Added new unittests to make check

Merge pull request #7574 from SUSE/wip-spec-license

ceph.spec.in: add license declaration

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

ceph.spec.in: add license declaration

Signed-off-by: Nathan Cutler <ncutler@suse.com>

cmake: Added new unittests to make check

Added unittest_shunique_lock & unittest_mutex_debug
to make check, and made them link ALLOC_LIBS

Signed-off-by: Ali Maredia <amaredia@redhat.com>

Merge pull request #7569 from SUSE/wip-copyright2

ceph.spec.in: add copyright notice

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

ceph.spec.in: add copyright notice

http://tracker.ceph.com/issues/14694 Fixes: #14694

Signed-off-by: Nathan Cutler <ncutler@suse.com>

os/bluestore/BlockDevice: std::atomic_int

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

os/bluestore/BlockDevice: use std::mutex and std::condition_variable

Pull aio wait code into an IOContext method too.

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

Merge pull request #7513 from athanatos/wip-ceph-test-rados-2

ceph_test_rados: use less CPU

Merge pull request #5931 from ceph/wip-cxx11concurrency

librados: move to c++11 concurrency types

Reviewed-by: Matt Benjamin <mbenjami@redhat.com>
Reviewed-by: Casey Bodley <cbodley@redhat.com>

Merge branch 'print-blocked-op' of git://github.com/majianpeng/ceph

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

Merge pull request #6608 from esnyder/wip-6634

osd: remove unused SnapContext member snapc from MOSDSubOp message

Reviewed-by: Xiaoxi Chen <xiaoxi.chen@intel.com>
Reviewed-by: Sage Weil <sage@redhat.com>

Merge branch 'master' of git://github.com/joconcepts/ceph

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

Merge pull request #7181 from songbaisen/song5

crush: reply quickly from get_immediate_parent

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

Merge pull request #7437 from Ved-vampir/wip_zlib_plugin

common: add zlib compression plugin

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

Merge pull request #7527 from liewegas/wip-no-rgw-prestart

systemd/ceph-radosgw-prestart.sh: remove

Reviewed-by: Nathan Cutler <ncutler@suse.com>

Merge pull request #7547 from yuyuyu101/wip-14679

AsyncConnection: avoid debug log in cleanup_handler

crush/CrushTester: workaround a bug in boost::icl

the boost::icl in boost v1.55 has a bug when working with libc++,

> Assertion failed: (this->_map.find(inter_val) == this->_map.end()),
> function gap_insert, file
> /usr/local/include/boost/icl/interval_base_map.hpp, line 555.

see https://svn.boost.org/trac/boost/ticket/9501 and
    https://svn.boost.org/trac/boost/ticket/9987

the bug was fixed in boost v1.56. as a workaround we use the the
`set` from `boost::container` instead of the `std::set` from libc++.
we should drop this workaround once we bump the supported boost
version from 1.42 to 1.56.

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

Merge pull request #7554 from tchaikov/wip-ceph-disk-test

tests: ceph-disk.sh should compare with the resolved path

Reviewed-by: Loic Dachary <ldachary@redhat.com>

OSDMap: reset osd_primary_affinity shared_ptr when deepish_copy_from

Base on current logic, OSDMonitor may call create_pending and
encode_pending twice for the some epoch.

In encode_pending:

tmp.deepish_copy_from(osdmap);
tmp.apply_incremental(pending_inc);

This Op would change the tmp osd_primary_affinity, but the osd_primary_affinity
is declared as ceph::shared_ptr, so this would change the osdmap too. When this
round encode_pending is proposed fail. We may call encode_pending again, but the
osdmap is changed last round, so the pending_inc would be wrong.

Fixes: #14686
Signed-off-by: Xinze Chi <xinze@xsky.com>

Merge pull request #7551 from JevonQ/update-libcephfs-version

libcephfs: update LIBCEPHFS_VERSION to indicate the interface was changed

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

tests: ceph-disk.sh should compare with the resolved path

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

libcephfs: Update LIBCEPHFS_VERSION to indicate the interface was changed

Signed-off-by: Jevon Qiao <scaleqiao@gmail.com>

AsyncMessenger: remove unused variables

* remove AsyncMessenger::listen_sd
* remove `ee` in KqueueDriver::del_event()

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

crypto: remove unused Digest::sec_type

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

journal: remove unused ObjectPlayer::m_fetch_interval

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

librbd: remove unused C_DiffObject::m_request_num

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

librbd: replace atomic_t with std::atomic<int>

it fails to compile with clang, and turns out it should be
`ceph::atomic_t`, but I take this chance to replace it
with std::atomic<> to phase out atomic_t.

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

ECUtil: remove unused stripe_info_t::stripe_size

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

osd: init HeartbeatDispatcher using OSD->cct

its base class 'Dispatcher' is uninitialized when used here to access
'Dispatcher::cct', so we'd better use OSD::cct to initialize it instead.

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

Merge pull request #7549 from tchaikov/wip-fix-ceph-disk-tests

ceph-disk/test: fix test_prepare.py::TestPrepare tests

Reviewed-by: Loic Dachary <ldachary@redhat.com>

ceph-disk/test: fix test_prepare.py::TestPrepare tests

actually we are verifying if the variable is an instance of
specified class. for example, the `prepare.data` should be
a `PrepareFilestoreData` if `--bluestore` is not specified.

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

AsyncConnection: avoid debug log in cleanup_handler

local connection will be stop and call cleanup_handler after messenger
is down
introduced in
commit(https://github.com/ceph/ceph/commit/9da2fffd31562ed5d0b795d7862b3ebec66aba40)

Fix #14679

Signed-off-by: Haomai Wang <haomai@xsky.com>

Merge pull request #7540 from dillaman/wip-librados-test-stub-unwatch

librados_test_stub: protect against notify/unwatch race

Reviewed-by: Josh Durgin <jdurgin@redhat.com>

Merge pull request #7529 from dillaman/wip-14412

librbd: journal framework for tracking exclusive lock transitions

Reviewed-by: Josh Durgin <jdurgin@redhat.com>

Merge pull request #7377 from liewegas/wip-datadir-search

config: add $data_dir/config to config search path

Merge pull request #7466 from adamemerson/wip-bsd-time-second-try

common: time: have skewing-now call non-skewing now

Merge pull request #7463 from dachary/wip-13422-pid-file

global: do not start two daemons with a single pid-file (part 2)

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

Merge pull request #7460 from xiexingguo/xxg-wip-block

osd: blockdevice: avoid implicit cast and add guard

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

Merge pull request #7530 from xiexingguo/xxg-wip-nid

osd: bluestore: assign nid for newly created onode

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

Merge pull request #7521 from liewegas/wip-memstore

osd: memstore: set Collection::cid on create

Merge pull request #7486 from ceph/wip-bluestore-oref

osd: bluestore: fix two wal bugs; reuse OnodeRefs when applying txc

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

Merge pull request #7531 from jjhuo/master

osd: bluestore: remove unused intrusive member hook in enode

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

Merge pull request #7535 from liewegas/wip-bluefs-mkfs

osd: bluestore/BlueFS: initialize super block_size earlier in mkfs

Reviewed-by: Dan Mick <dmick@redhat.com>

Merge pull request #7536 from liewegas/wip-bluestore-block

osd: bluestore: fix block device file creation

Merge pull request #7541 from linuxbox2/cmake-feb5

Cmake feb5

osdc: Update to use C++11 concurrency

The only externally visible change is that Objecter::get_osdmap_read
and Objedcter::put_osdmap_read have been removed in favor of
Objecter::with_osdmap.

It can be used like:

objecter->with_osdmap([&](const OSDMap& o) { o.do_stuff(); };
int x = objecter->with_osdmap([&](const OSDMap& o) {
    return get_thing(id); });
objecter->with_osdmap(std::mem_fn(&OSDMap::do_stuff));
int x = objecter->with_osdmap(std::mem_fn(&OSDMap::get_thing),
                              id);

The choice between the style of the first two examples or the second two
is arbitrary and depends on what you prefer.

Signed-off-by: Adam C. Emerson <aemerson@redhat.com>