| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| |
|
|
|
| |
Change-Id: Ia84cc24c8924e6d22d02ac15f611c10e26db99b4
Signed-off-by: Nigel Babu <nigelb@redhat.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The state management of "connected" in rpc is ad-hoc as far as the
responsibility goes. Note that there is nothing wrong with
functionality itself. rpc layer manages this state in disconnect
codepath and has exposed an api to manage this one from
consumers. Note that rpc layer never sets "connected" to true by
itself, which forces the consumers to use this api to get a working
rpc connection. The situation is best captured from a comment in code
from Jeff Darcy in glusterfsd/src/gf-attach.c:
-/*
- * In a sane world, the generic RPC layer would be capable of tracking
- * connection status by itself, with no help from us. It might invoke our
- * callback if we had registered one, but only to provide information. Sadly,
- * we don't live in that world. Instead, the callback *must* exist and *must*
- * call rpc_clnt_{set,unset}_connected, because that's the only way those
- * fields get set (with RPC both above and below us on the stack). If we don't
- * do that, then rpc_clnt_submit doesn't think we're connected even when we
- * are. It calls the socket code to reconnect, but the socket code tracks this
- * stuff in a sane way so it knows we're connected and returns EINPROGRESS.
- * Then we're stuck, connected but unable to use the connection. To make it
- * work, we define and register this trivial callback.
- */
Also, consumers of rpc know about state of connection only through the
notifications sent by rpc-clnt. So, consumers don't have any extra
information to manage the state and hence letting them manage the
state is counter intuitive. This patch cleans that up and instead
moves the responsibility of state management of rpc layer into
itself.
Change-Id: I31e641a60795fc480ca753917f4b2579f1e05094
Signed-off-by: Raghavendra G <rgowdapp@redhat.com>
Fixes: bz#1585585
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Summary:
Halo Geo-replication is a feature which allows Gluster or NFS clients to write
locally to their region (as defined by a latency "halo" or threshold if you
like), and have their writes asynchronously propagate from their origin to the
rest of the cluster. Clients can also write synchronously to the cluster
simply by specifying a halo-latency which is very large (e.g. 10seconds) which
will include all bricks.
In other words, it allows clients to decide at mount time if they desire
synchronous or asynchronous IO into a cluster and the cluster can support both
of these modes to any number of clients simultaneously.
There are a few new volume options due to this feature:
halo-shd-latency: The threshold below which self-heal daemons will
consider children (bricks) connected.
halo-nfsd-latency: The threshold below which NFS daemons will consider
children (bricks) connected.
halo-latency: The threshold below which all other clients will
consider children (bricks) connected.
halo-min-replicas: The minimum number of replicas which are to
be enforced regardless of latency specified in the above 3 options.
If the number of children falls below this threshold the next
best (chosen by latency) shall be swapped in.
New FUSE mount options:
halo-latency & halo-min-replicas: As descripted above.
This feature combined with multi-threaded SHD support (D1271745) results in
some pretty cool geo-replication possibilities.
Operational Notes:
- Global consistency is gaurenteed for synchronous clients, this is provided by
the existing entry-locking mechanism.
- Asynchronous clients on the other hand and merely consistent to their region.
Writes & deletes will be protected via entry-locks as usual preventing
concurrent writes into files which are undergoing replication. Read operations
on the other hand should never block.
- Writes are allowed from _any_ region and propagated from the origin to all
other regions. The take away from this is care should be taken to ensure
multiple writers do not write the same files resulting in a gfid split-brain
which will require resolution via split-brain policies (majority, mtime &
size). Recommended method for preventing this is using the nfs-auth feature to
define which region for each share has RW permissions, tiers not in the origin
region should have RO perms.
TODO:
- Synchronous clients (including the SHD) should choose clients from their own
region as preferred sources for reads. Most of the plumbing is in place for
this via the child_latency array.
- Better GFID split brain handling & better dent type split brain handling
(i.e. create a trash can and move the offending files into it).
- Tagging in addition to latency as a means of defining which children you wish
to synchronously write to
Test Plan:
- The usual suspects, clang, gcc w/ address sanitizer & valgrind
- Prove tests
Reviewers: jackl, dph, cjh, meyering
Reviewed By: meyering
Subscribers: ethanr
Differential Revision: https://phabricator.fb.com/D1272053
Tasks: 4117827
Change-Id: I694a9ab429722da538da171ec528406e77b5e6d1
BUG: 1428061
Signed-off-by: Kevin Vigor <kvigor@fb.com>
Reviewed-on: http://review.gluster.org/16099
Reviewed-on: https://review.gluster.org/16177
Tested-by: Pranith Kumar Karampuri <pkarampu@redhat.com>
Smoke: Gluster Build System <jenkins@build.gluster.org>
NetBSD-regression: NetBSD Build System <jenkins@build.gluster.org>
CentOS-regression: Gluster Build System <jenkins@build.gluster.org>
Reviewed-by: Pranith Kumar Karampuri <pkarampu@redhat.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
http://review.gluster.org/14085 fixes a/the "leak" - via the
generated rpc/xdr headers - of pragmas that mask these warnings.
However 14085 won't pass the smoke test until all the warnings are
fixed.
Change-Id: I3d8a7a3de35058aa97eab59d3f59208396298b03
BUG: 1369124
Signed-off-by: Kaleb S. KEITHLEY <kkeithle@redhat.com>
Reviewed-on: http://review.gluster.org/15246
Smoke: Gluster Build System <jenkins@build.gluster.org>
NetBSD-regression: NetBSD Build System <jenkins@build.gluster.org>
CentOS-regression: Gluster Build System <jenkins@build.gluster.org>
Reviewed-by: Kotresh HR <khiremat@redhat.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
[WIP patch as of now, just needs a little tweak]
A pending TODO in the code caused regressions to fail as
bitrot daemons are spawned during volume start (equivalent
to enabling bitrot by default). The problematic part that
casued such failures is during brick disconnections with
unsafe handling of event data structured in the code.
With this patch, data structures are properly cleaned up
with care taken to cleanup all accessors first. This also
fixes potential memory leaks which was bluntly ignored
before.
Change-Id: I70ed82cb1a0fb56c85ef390007e321a97a35c5ce
BUG: 1170075
Signed-off-by: Venky Shankar <vshankar@redhat.com>
original-author: Venky Shankar <vshankar@redhat.com>
Reviewed-on: http://review.gluster.org/9959
Reviewed-by: Vijay Bellur <vbellur@redhat.com>
Tested-by: Vijay Bellur <vbellur@redhat.com>
|
|
|
This patch introduces RPC based communication between the changelog
translator and libgfchangelog. It replaces the old pathetic stream
based interaction that existed earlier (due to time constraints :-/).
Changelog, upon initialization starts a RPC server (rpcsvc) allowing
clients to invoke a probe API as a bootup mechanism to request for
event notifications. During probe, clients can choose an event
filter specifying the type(s) of events they are interested in. As
of now there is no way to change the event notification set once
the probe RPC call is made, but that is easier to implement.
The actual event notifications is done on a separate RPC session.
The client (libgfchangelog) itself starts and RPC server which the
changelog translator "connects back" during probe. Notifications
are dispatched by a bunch of threads from the server (translator)
and the client optionally orders them if ordered notifications
are requried. FOPs fill in their respective event details in a
buffer (rot-buffs to be particular) and a bunch of threads
(consumers) swap the buffers out of roatation and dispatch them
via RPC. To avoid writer starvation, then number of dispatcher
threads is one less than the number of buffer list in rot-buffs.x
libgfchangelog becomes purely callback based -- upon event
notification from the server (and re-ordering them if required)
invoke a callback routine specified by consumer(s).
A major part of the patch is also aimed at providing backward
compatibility for geo-replication, which was one of the main
consumer of the stream based API. Also, this patch does not\
"turn on" event notifications for all fops, just a bunch which
is currently in requirement. Another pain point is that the
server does not filter events before dispatching it to the
clients. That load is taken up by the client itself (although
it's done at the library layer rather than making it hard on
the callback implementor). This needs improvement and care
needs to be taken to not load the server up with expensive
filtering mechanisms.
Change-Id: Ibf60a432b68f2dfa60c6f9add2bcfd37a9c41395
BUG: 1170075
Signed-off-by: Venky Shankar <vshankar@redhat.com>
Reviewed-on: http://review.gluster.org/9708
Reviewed-by: Jeff Darcy <jdarcy@redhat.com>
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Vijay Bellur <vbellur@redhat.com>
|
