This is based of @neilalexander PR #3558.
It ensures that the timer is reset/canceled on configuration
update (by the leader only).
Fixed also the issue with a super-cluster where the delete timer
would always be reset at every gateway interval check.
Signed-off-by: Ivan Kozlovic <ivan@synadia.com>
If a stream is created or updated with a negative replicas count,
and error is now returned. Same for consumers.
Signed-off-by: Ivan Kozlovic <ivan@synadia.com>
It could be that while the routine processing the consumer assignment
runs the stream is being stopped, which would lead to a panic.
Signed-off-by: Ivan Kozlovic <ivan@synadia.com>
A request to `$SYS.REQ.SERVER.PING.JSZ` would now return something
like this:
```
...
"meta_cluster": {
"name": "local",
"leader": "A",
"peer": "NUmM6cRx",
"replicas": [
{
"name": "B",
"current": true,
"active": 690369000,
"peer": "b2oh2L6w"
},
{
"name": "Server name unknown at this time (peerID: jZ6RvVRH)",
"current": false,
"offline": true,
"active": 0,
"peer": "jZ6RvVRH"
}
],
"cluster_size": 3
}
```
Note the "peer" field following the "leader" field that contains
the server name. The new field is the node ID, which is a hash of
the server name.
Signed-off-by: Ivan Kozlovic <ivan@synadia.com>
If the client pull requests has a max_bytes value and the server
cannot deliver a single message (because size is too big), it
is sending a 409 to signal that to the client library. However,
if it sends at least a message then it would close the request
without notifying the client with a 409, which would cause the
client library to have to wait for its expiration/timeout.
Signed-off-by: Ivan Kozlovic <ivan@synadia.com>
This fixes an issue introduced in #3080
The consumer filter subject check was skipped on recovery.
The intent was to bypass the upstream stream subjects.
But it also filtered the downstream stream subject.
This became a problem when the downstream was itself an upstream.
Then during recover, the stream subject was not checked, which
lead to delivery of filtered messages that should never have been
delivered.
Signed-off-by: Matthias Hanel <mh@synadia.com>
- didRemove in applyMetaEntries() could be reset when processing
multiple entries
- change "no race" test names to include JetStream
- separate raft nodes leader stepdown and stop in server
shutdown process
- in InstallSnapshot, call wal.Compact() with lastIndex+1
Signed-off-by: Ivan Kozlovic <ivan@synadia.com>
This is a regression introduced in v2.8.3. If a message reaches
the max redeliver count, it stops being delivered to the consumer.
However, after a server or cluster restart, those messages would
be redelivered again.
Resolves#3361
Signed-off-by: Ivan Kozlovic <ivan@synadia.com>
The Move/Cancel/Downscale mechanism did not take into account that
the consumer's replica count can be set independently.
This also alters peer selection to have the ability to skip
unique tag prefix check for server that will be replaced.
Say you have 3 az, and want to add another server to az:1,
in order to replace a server that is the same zone.
Without this change, uniqueTagPrefix check would filter
the server to replace with and cause a failure.
The cancel move response could not be received due to
the wrong account name.
Signed-off-by: Matthias Hanel <mh@synadia.com>
Make sure when processing a peer removal that the stream assignment agrees.
When a new leader takes over it can resend a peer removal, and if the stream/consumer really was rescheduled we could remove by accident.
Also need to make sure that when we remove a stream we remove the node as part of the stream assignment.
If we didn't, if the same asset returned to this server we would not start up the monitoring loop.
Simplify migration logic in monitorStream, to be driven by leader only
Improved unit tests
Added failure when server not in peer list
Move command does not require server anymore
Signed-off-by: Matthias Hanel <mh@synadia.com>
This effectively means that requests with batch > 1 will process a message and go to the end of the line.
Signed-off-by: Derek Collison <derek@nats.io>
When creating a pull consumer with InactiveThreshold set, if the
application is doing pull requests with "no_wait" at regular interval
(lower than InactiveThreshold), the JS consumer should be considered
active and not deleted.
Signed-off-by: Ivan Kozlovic <ivan@synadia.com>
Also for direct get and for pull requests, if we are not on a client connection check how long we have been away from the readloop.
If need be execute in a separate go routine.
Signed-off-by: Derek Collison <derek@nats.io>
This could happen when a consumer had not sent anything to the
attached NATS subscription and there was a consumer leader
step down or server restart.
Signed-off-by: Derek Collison <derek@nats.io>
Before PR #3099, `waitQueue.isEmpty()` returned `wq.len() == 0`
and `waitQueue.len()` was protecting against the pointer being
nil (and then return 0).
The change in #3099 caused `waitQueue.isEmpty()` to return `wq.n == 0`,
which means that if `wq` was nil, then it would crash.
This PR restores `waitQueue.isEmpty()` to return `wq.len() == 0` and
add the protection for waitQueue being nil in `len()` similar to
how it was prior to PR #3099.
Resolves#3117
Signed-off-by: Ivan Kozlovic <ivan@synadia.com>
There was an issue with MaxWaiting==1 that was causing a request
with expiration to actually not expire. This was because processWaiting
would not pick it up because wq.rp was actually equal to wq.wp
(that is, the read pointer was equal to write pointer for a slice
of capacity of 1).
The other issue was that when reaching the maximum of waiting pull
requests, a new request would evict an old one with a "408 Request Canceled".
There is no reason for that, instead the server will first try to
find some existing expired requests (since some of the expiration
is lazily done), but if none is expired, and the queue is full,
the server will return a "409 Exceeded MaxWaiting" to the new
request, and not a "408 Request Canceled" to an old one...
Signed-off-by: Ivan Kozlovic <ivan@synadia.com>
When updating usage, there is a lock inversion in that the jetStream
lock was acquired while under the stream's (mset) lock, which is
not correct. Also, updateUsage was locking the jsAccount lock, which
again, is not really correct since jsAccount contains streams, so
it should be jsAccount->stream, not the other way around.
Removed the locking of jetStream to check for clustered state since
js.clustered is immutable.
Replaced using jsAccount lock to update usage with a dedicated lock.
Originally moved all the update/limit fields in jsAccount to new
structure to make sure that I would see all code that is updating
or reading those fields, and also all functions so that I could
make sure that I use the new lock when calling these. Once that
works was done, and to reduce code changes, I put the fields back
into jsAccount (although I grouped them under the new usageMu mutex
field).
Signed-off-by: Ivan Kozlovic <ivan@synadia.com>
When deciding to compact a file, we need to remove from the raw
bytes the empty records, otherwise, for small messages, we would
end-up calling compact() too many times.
When removing a message from the stream, in FIFO cases we would
write the index every 2 seconds at most when doing it in place,
when when dealing with out of order deletes, we would do it for
every single delete, which can be costly. We are now writing
only every 500ms for non FIFO cases.
Also fixed some unrelated code:
- Decision to install a snapshot was based on incorrect logical
expression
- In checkPending(), protect against the timer being nil which
could happen when consumer is stopped or leadership change.
Signed-off-by: Ivan Kozlovic <ivan@synadia.com>
Step down timing for consumers or streams.
Signals loss of leadership and sleeps before stepping down.
This makes it less likely that messages are being processed during step
down.
When becoming leader, consumer stream seqno got reset,
even though the consumer existed already.
Proper cleanup of redelivery data structures and timer
Signed-off-by: Matthias Hanel <mh@synadia.com>
