mds: limit size of MExportDir message

Only export port of subtree if whole subtree tree is too large for
sinlge MExportDir message. The un-exported portion are added to
export queue.

Fixes: http://tracker.ceph.com/issues/21892
Signed-off-by: "Yan, Zheng" <zyan@redhat.com>

mds: optimize MDCache::try_subtree_merge

Signed-off-by: "Yan, Zheng" <zyan@redhat.com>

mds: optimize import/export state access

Signed-off-by: "Yan, Zheng" <zyan@redhat.com>

Merge pull request #17983 from liewegas/wip-balancer

mgr/balancer: make crush-compat mode work

Reviewed-by: John Spray <john.spray@redhat.com>

Merge pull request #18657 from cbodley/wip-21990

rgw: remove placement_rule from rgw_link_bucket()

Merge pull request #18534 from ZVampirEM77/wip-emlc-contentmd5

rgw: lc support Content-MD5 request header and fix a rgw crash bug

Reviewed-by: Casey Bodley <cbodley@redhat.com>
Reviewed-by: Abhishek Lekshmanan <abhishek.lekshmanan@gmail.com>
Reviewed-by: Adam Emerson <aemerson@redhat.com>
Reviewed-by: Daniel Gryniewicz <dang@redhat.com>

Merge pull request #18610 from cbodley/wip-osd-subop-num

osd: add processed_subop_count for cls_cxx_subop_version()

Reviewed-by: Jason Dillaman <dillaman@redhat.com>
Reviewed-by: Sage Weil <sage@redhat.com>
Reviewed-by: Josh Durgin <jdurgin@redhat.com>

Merge pull request #18641 from joke-lee/rgw_modify_subuser_access_error

rgw: modify s3 type subuser access permissions fail through admin rest api

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

Merge pull request #18659 from tchaikov/wip-21993

qa: stop testing deprecated "ceph osd create"

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

Merge pull request #15193 from mynaramana/arraycode

osd: introduce sub-chunks to erasure code plugin interface

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

Merge pull request #18637 from yaozongyou/fix-dashboard-bug

osd: make stat_bytes and stat_bytes_used counters PRIO_USEFUL

Reviewed-by: Kefu Chai <kchai@redhat.com>
Reviewed-by: John Spray <john.spray@redhat.com>

Merge pull request #18347 from Synesis-LLC/fix-21773

mgr: skip first non-zero incremental in PGMap::apply_incremental()

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

Merge pull request #18453 from liewegas/wip-osd-devs

osd: enumerate device names in a simple way

Reviewed-by: John Spray <john.spray@redhat.com>

Merge pull request #18509 from tchaikov/wip-osd-cleanups

osd,mgrclient: pass daemon_status by rvalue ref and other cleanups

Reviewed-by: John Spray <john.spray@redhat.com>

rgw: remove placement_rule from cls_user_bucket_entry

placement_rule is no longer needed in cls_user_bucket_entry, because the
only time that it's needed, we can read it from the bucket instance in
RGWRados::update_containers_stats()

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

Merge pull request #18504 from badone/wip-GroupSnapshotNamespace-group_pool-uninitialised

cls/rbd: Silence gcc7 maybe-uninitialized warning

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

rgw: remove placement_rule from rgw_link_bucket()

with the fallback in RGWRados::update_containers_stats(), we no longer
need to pass placement_rule into the cls_user_bucket_entry

this removes the dependency between bucket entrypoint metadata and its
bucket instance metadata during multisite sync

Fixes: http://tracker.ceph.com/issues/21990
Signed-off-by: Casey Bodley <cbodley@redhat.com>

rgw: take placement_rule from bucket info in update_containers_stats

in case the cls_user_bucket_entry doesn't contain a placement_rule, take
it from the bucket instance info

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

mgr/balancer: simplify pool_info tracking

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

mgr/balancer: less verbose on 'eval' by default; add 'eval-verbose'

The verbose output is helpful for debugging and understanding what is
being measured, but most of the time all that you care about is the score.

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

mgr/balancer: fix pg vs object terminology

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

mgr/balancer: restrict to time of day

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

mgr/module: adjust osd_weight min step to .005

That should be ~1 PG or less on average; no real sense it taking a
step smaller than that!

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

mgr/balancer: if score regresses, take a few more steps

To avoid getting stuck in a local minima, take a few additional
steps if our score drops and see if it gets better.

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

mgr/balancer: allow 5% misplaced

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

mgr/balancer: more aggressive steps

We are smart enough to back off if we misplace too much.

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

qa/suites/rados/thrash/d-balancer: enable balancer in various modes

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

mgr/balancer: crush-compat: phase out osd_weights

Phase out the osd_weight values as we optimize the crush compat
weight set.  Allow a small regression in overall score if it means
we are taking a step to remove the osd_weight value.

Fix the 'osd reweightn' invocation.

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

mgr/balancer: crush_compat: cope with 'out' osds

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

mgr/balancer: stop if we get a perfect score

..and fix the 'worse' threshold (we're just working around floating
point imprecision; 1.01 was too coarse).

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

mgr/balancer: more dead code

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

mgr/balancer: crush-compat: throttle changes based on max_misplaced

Take smaller steps if we overshoot max_misplaced.

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

mgr/balancer: remove dead code

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

mgr/balancer: include pg up mapping in MappingState

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

mgr/balancer: normalize weight-set weights to sum to target weight

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

mgr/balancer: note root id in Eval

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

mgr/balancer: make crush-compat mode work!

- it does multiple iterations, like the upmap optimizer.
- it decreases the step size if it isn't improving, in the hope
that it is overshooting the minimum
- debug output is cleaned up a bit (the info level should be
genuinely useful)

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

Merge pull request #16651 from jcsp/wip-mgr-module-interface

mgr: python interface rework + enable modules to run in standby mode

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

mon: include disabled modules in `mgr module ls`

Otherwise, when someone wants to see what's possible
to do with `mgr module enable` they have to trawl
through the whole mgr map dump.

Signed-off-by: John Spray <john.spray@redhat.com>

doc: describe using `mgr module ...` commands

...including the new "mgr services" command.

Signed-off-by: John Spray <john.spray@redhat.com>

doc: describe how to implement standby modules

Signed-off-by: John Spray <john.spray@redhat.com>

qa: fix mgr caps

This was still using Kraken era settings

Signed-off-by: John Spray <john.spray@redhat.com>

qa: expand mgr testing

Some extra coverage of the dashboard, including its standby
redirect mode and the publishing of URIs.

Also invoking the command_spam mode of the selftest module.

Signed-off-by: John Spray <john.spray@redhat.com>

mgr/selftest: extend test and add background spam mode

Signed-off-by: John Spray <john.spray@redhat.com>

mgr: drop GIL around set_uri, set_health_checks

These didn't need to keep the GIL to go and do their
pure C++ parts, and by keeping it they could deadlock
while trying to take ActiveMgrModules::lock.

Signed-off-by: John Spray <john.spray@redhat.com>

mgr: fix ~MonCommandCompletion

This was doing a Py_DECREF outside of the Gil.

Fixes: http://tracker.ceph.com/issues/21593
Signed-off-by: John Spray <john.spray@redhat.com>

mgr: update for SafeThreadState

A bunch of the previous commits were done
before this class existed, so updating in
one go instead of trying to edit history
in fine detail.

Signed-off-by: John Spray <john.spray@redhat.com>

mgr: refactor PyOSDMap etc implementation

Implement real python classes from the C side,
rather than exposing only module methods.

Signed-off-by: John Spray <john.spray@redhat.com>

mgr/PyOSDMap: add CRUSH get_item_weight

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

mgr: fix py_module_registry shutdown

Was calling way too early, which did a
Py_Finalize before the modules had been
joined.

Signed-off-by: John Spray <john.spray@redhat.com>

mgr: fix thread naming

Was passing a reference to a local stringstream into
Thread::create, not realising that it was taking a char*
reference instead of a copy.  Result was garbage (or usually,
all threads having the name of the last one created)

Signed-off-by: John Spray <john.spray@redhat.com>

mgr: cut down duplication between active+standby

...by using PyModuleRunner class from ActivePyModule too.

Signed-off-by: John Spray <john.spray@redhat.com>

mgr: fix os._exit overrides

These would throw an exception when passed
a status code.

Signed-off-by: John Spray <john.spray@redhat.com>

mon/MgrMonitor: reset services map on drop_active

Signed-off-by: John Spray <john.spray@redhat.com>

mgr/dashboard: implement standby mode

Signed-off-by: John Spray <john.spray@redhat.com>

pybind/mgr: add MgrStandbyModule

Signed-off-by: John Spray <john.spray@redhat.com>

mgr: standby modules come up and run now

...they still don't have access to any config though.

Signed-off-by: John Spray <john.spray@redhat.com>

mgr: enable running modules in standby mode

Modules can implement a second, separate class
that has access to very little state about the
system and can't implement commands.

They have just enough information to redirect
or forward incoming requests/traffic to the
active instance of the module on the active mgr.

This enables module authors to create modules
that end users can access via any (running) mgr node
at any time, rather than having to first work out
which mgr node is active.

Signed-off-by: John Spray <john.spray@redhat.com>

mgr: clean up python source file naming

Signed-off-by: John Spray <john.spray@redhat.com>

mgr: refactor python module management

Separate out the *loading* of modules from
the *running* of modules.

This is a precursor to enabling modules to run
in standby mode.

Signed-off-by: John Spray <john.spray@redhat.com>

pybind/mgr: use set_uri hook from dashboard+restful modules

No more guessing the URL!

Signed-off-by: John Spray <john.spray@redhat.com>

mgr: enable python modules to advertise their service URI

Fixes: http://tracker.ceph.com/issues/17460
Signed-off-by: John Spray <john.spray@redhat.com>

mon/MgrMonitor: store services in map and expose with command

Signed-off-by: John Spray <john.spray@redhat.com>

messages: `services` in MMgrBeacon

Signed-off-by: John Spray <john.spray@redhat.com>

mon/MgrMap: store list of services

Signed-off-by: John Spray <john.spray@redhat.com>

mgr: carry PyModules ref in MonCommandCompletion

Signed-off-by: John Spray <john.spray@redhat.com>

pybind: update MgrModule for ceph_state->ceph_module

& tidy up the places where ceph_state was getting
used outside of MgrModule.

Signed-off-by: John Spray <john.spray@redhat.com>

mgr: refactor python interface

Expose a python class instead of a module,
so that we have a place to carry our reference
to our MgrPyModule* and to PyModules*, rather than
passing a handle for the former and using
a global pointer for the latter.

Signed-off-by: John Spray <john.spray@redhat.com>

mgr/dashboard: remove blue highlight on scrubbing pg states

This was kind of unnecessary, highlighting a completely normal
and healthy situation in a different colour.  The blue was
also really hard to read against a grey background.

Signed-off-by: John Spray <john.spray@redhat.com>

mgr/dashboard: clean up fs standby list when empty

Signed-off-by: John Spray <john.spray@redhat.com>

Merge pull request #18243 from jan--f/mgr-prom-osd-status

pybind/mgr/prometheus: add osd_in/out metric; make osd_weight a metric

Reviewed-by: John Spray <john.spray@redhat.com>

Merge pull request #18651 from aermolaev/patch-1

doc: fix typo in radosgw/dynamicresharding.rst

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

doc: fix typo in dynamicresharding.rst: admon => admin

Signed-off-by: Alexander Ermolaev <ave@integros.com>

mgr: skip first non-zero incremental in PGMap::apply_incremental()

After initialization of PGMap instance PGMap::stamp is zero
and this cause huge first delta.
Also after mgr restart first non-zero value coming to PGMap::apply_incremental()
is current pg_sum value so it produces unreasonably huge pg_sum_delta.
This patch introduces a workaround to save pg_sum and not update pg_sum_delta
by first non-zero incremental.

Signed-off-by: Aleksei Gutikov <aleksey.gutikov@synesis.ru>
Fixes: http://tracker.ceph.com/issues/21773

Merge pull request #18660 from ownedu/wip-fix-async-rdma-typo

msg/async/rdma: a tiny typo fix.

Reviewed-by: Kefu Chai <kefu@redhat.com>
Reviewed-by: Haomai Wang <haomai@xsky.com>

Merge pull request #18268 from jcsp/wip-20629

mon: handle monitor lag when killing mgrs

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

cls_rbd: Silence gcc7 maybe-uninitialized warning

gcc7 warns that GroupSnapshotNamespace::group_pool may be used
uninitialized in SnapshotNamespaceOnDisk::decode. Reordering the member
variables silences the spurious warning.

Signed-off-by: Brad Hubbard <bhubbard@redhat.com>

msg/async/rdma: a tiny typo fix.

Signed-off-by: Yan Lei <yongyou.yl@alibaba-inc.com>

Merge pull request #18329 from ifed01/wip-fix-mempool-dump

mempool: fix lack of pool names in mempool:dump output for JSON format

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

Merge pull request #18575 from majianpeng/msg-cleanup-2

msg/async:  misc cleanup.

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

Merge pull request #18581 from wzliao/wip-drop-unused-parameter

osd/PrimaryLogPG: drop unused parameters

Reviewed-by: Jos Collin <jcollin@redhat.com>

Merge pull request #18588 from mogeb/rados-bench-clock

tools/rados: use the monotonic clock in rados bench

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

Merge pull request #18595 from batrick/mon-clock

mon: a few conversions to monotonic clock

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

Merge pull request #18636 from tchaikov/wip-osdmap-cleanup

osd/OSDMap: remove the unnecessary checks for null

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

Merge pull request #18647 from tchaikov/wip-pgmap-cleanup

mon/PGMap: use new-style options and cleanup

Reviewed-by: John Spray <john.spray@redhat.com>

Merge pull request #18648 from tchaikov/wip-vstart-drop-num-opts

vstart.sh: drop --{mon,osd,mds,rgw,mgr}_num options

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

qa: stop testing deprecated "ceph osd create"

"ceph osd create" is not idempotent, and is considered deprecated.

Fixes: http://tracker.ceph.com/issues/21993
Signed-off-by: Kefu Chai <kchai@redhat.com>

Merge PR #18618 into master

* refs/pull/18618/head:
mds: pass a reference of function to sanitize() not a copy

Reviewed-by: Patrick Donnelly <pdonnell@redhat.com>
Reviewed-by: Zheng Yan <zyan@redhat.com>
Reviewed-by: Amit Kumar <amitkuma@redhat.com>

Merge pull request #18572 from linuxbox2/wip-rgwfile-env

rgw_log, rgw_file: account for new required envvars

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

Merge pull request #18327 from PCzhangPC/clonelostmeta

librbd: cannot clone all image-metas if we have more than 64 key/value pairs

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

Merge pull request #18546 from liewegas/wip-deprecate-features

include/ceph_features: deprecate a bunch of features

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

Merge pull request #18616 from xiexingguo/wip-hunt-scrub-mismatch

osd/PrimaryLogPG: more oi.extents fixes

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

Merge pull request #18634 from tchaikov/wip-qa-add-restful

qa: add "restful" to ceph_mgr_modules in ceph-ansible suite

os/filestore: implement get_devices()

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

os/bluestore: implement get_devices()

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

osd: return rolled-up 'devices' list as part of metadata

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

os/bluestore/BlockDevice: add get_devname(), get_devices()

Enumerate device(s) that we are backed by.  It may be plural because of
DM, which might mask an underlying device, or because the dm device is
combining multiple other devices.

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

Merge pull request #18520 from alram/kube-helm-doc

doc: Kube + Helm installation

os/ObjectStore: add get_devices() method

The intent here is to return the set of hardware storage devices, as
identified by the "devname" (i.e., the directory name for the device in
/sys/block/).  For example, "sda", "nvme0n1", etc.

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

Merge pull request #18639 from yaozongyou/make-readme-properly-align

doc: make the commands in README.md properly aligned

Reviewed-by: Amit Kumar <amitkuma@redhat.com>