If write timeout and last acknowledgement needed for CL happen simultaneously
_ready can be sent to be exceptional by the timeout handler, but since
removal of the response handler happens in continuation it may be
reordered with last ack processing and there _ready will be set again
which will trigger assert. Fix it by removing the handler immediately,
no need to wait for continuation. It makes code simpler too.
The get_cm_stats gets a pointer to a field in the stats object. It
should capture it by value or segmentation falut may occure when the
caller gets out of scope.
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
Currently, we don't let the user know even what is the filename that failed.
That information should be included in the message.
Signed-off-by: Glauber Costa <glommer@scylladb.com>
This assert (in write fiber) would fail if read fiber failed
because the variable done will not be set to true.
The use of assert is very bad, because it prevents scylla
from proceeding, which is possible.
To solve it, let's trigger an exception if done is not true.
We do have code that will wait for both read and write fibers,
and catch exceptions, if any.
Closes#523.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Since 4641dfff24, query_state keeps a
copy of client_state, not a reference. Therefore _cl is no longer
updated by queries using _qp. Fix by using the client_state from _qp.
Fixes#525.
All responses sent from the server have protocol version set to
connection::_version which is set to the version used by the client
in its first message. However, if the protocol version used by the
client is unuspported or invalide the server should use the latest
version it recognizes.
This solves problem with version negotiation with Java driver. The
driver first sends a request in the latest version it recognizes, if
that fails it retries with the version that server has used in the error
message. If that fails as well it gives up. However, since Scylla always
responds with the same version that the client has used the negotiation
always fails if the client supports more protocol version than the
server.
Refs #317.
Signed-off-by: Paweł Dziepak <pdziepak@scylladb.com>
Get initial tokens specified by the initial_token in scylla.conf.
E.g.,
--initial-token "-1112521204969569328,1117992399013959838"
--initial-token "1117992399013959838"
It can be multiple tokens split by comma.
"This series adds the missing functionality that the nodetool describering would work.
It import the missing functionality from origin.
After this patch the API:
GET /storage_service/describe_ring/{keyspace}
will be available"
This patch chanages the API to support describe ring instead of describe
ring jmx that will be implemented in the jmx server.
The API will return a list of objects instead of string.
An additional api was added as the equivelent to the jmx call with an
empty param.
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
This patch adds the following methods implementation:
getRpcaddress
getRangeToAddressMap
getRangeToAddressMapInLocalDC
describeRing
getAllRanges
Those methods are used as part of the describe_ring method
implementation.
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
The storage server uses the token_range in origin to return inforamtion
about the ring.
This import the structures. The functionality in origin is redundant in
this case and was not imported.
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
Use all the disks except the one for rootfs for RAID0 which stores
scylla data. If only one disk is available warn the user since currently
our AMI's rootfs is not XFS.
[fedora@ip-172-31-39-189 ~]$ cat WARN.TXT
WARN: Scylla is not using XFS to store data. Performance will suffer.
Tested on AWS with 1 disk, 2 disks, 7 disk case.
(cherry picked from commit 49d6cba471)
Mistakenly didn't included on yum repository for AMI patchset, but it's needed
Signed-off-by: Takuya ASADA <syuu@cloudius-systems.com>
(cherry picked from commit 8587c4d6b3)
The nodetool cleanup command is used in many of the tests, because the
API call is not implemented it causes the tests to fail.
This is a workaround until the cleanup will be implemented, the method
return successfuly.
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
Normally an API call that is not implemented should fail, there are
cases that as a workaround an API call is stub, in those cases a warning
is added to indicate that the API is not implemented.
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
This patch do the following:
It adds a getter for the completed respond messages (i.e. the total
messages that were sent by the server)
It replaces the return mapping for the statistics to use the key, value
notation that is used in the jmx side.
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
This adds the read repair statistics to he storage_proxy stats and adds
to its implementation incrementing the counters value.
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
The API needs to get the stats from the rpc server, that is hidden from the
messaging service API.
This patch adds a foreach function that goes over all the server stats
without exposing the server implementation.
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
"The main objective of the series is to introduce statistics about ongoing
read/writes and especially those that are done in the background (acknowledged,
but uncompleted), but it contains some cleanups as well."
Add statistics for ongoing reads and ongoing background reads. Read is
a background one if it was acknowledged, but there still work to do to
complete it.
