If a stream failed, print a clear error message that repair failed, instead
of ignoring it and letting Seastar's generic "warning, exception was ignored"
be the only thing the user will see.
Signed-off-by: Nadav Har'El <nyh@cloudius-systems.com>
The previous repair code exchanged data with the other nodes which have
one arbitrary token. This will only work correctly when all the nodes
replicate all the data. In a more realistic scenario, the node being
repaired holds copies of several token ranges, and each of these ranges
has a different set of replicas we need to perform the repair with.
So this patch does the right thing - we perform a separate repair_range()
for each of the local ranges, and each of those will find a (possibly)
different set of nodes to communicate with.
Signed-off-by: Nadav Har'El <nyh@cloudius-systems.com>
This patch adds a method get_ranges() to replication-strategy.
It returns the list of token ranges held by the given endpoint.
It will be used by the replication code, which needs to know
in particular which token ranges are held by *this* node.
This function is the analogue of Origin's getAddressRanges().get(endpoint).
As in Origin, also here the implementation is not meant to be efficient,
and will not be used in the fast path.
Signed-off-by: Nadav Har'El <nyh@cloudius-systems.com>
Now, make_local_reader does not need partition_range to be alive when we
read the mutation reader. No need to store it in stream_detail for its
lifetime.
Since all the gossip callback (e.g., on_change) are executed inside a
seastar::async context, we can make wait for the operations like update
system table to complete.
So that do_before_change_notifications and do_on_change_notifications
are under seastar::async.
Now, before_change callbacks are inside seastar::async context.
It is easier to futurize apply_new_states and handle_major_state_change.
Now, on_change, on_join and on_restart callbacks are inside
seastar::async context.
It is not correct to use _scheduled_gossip_task.armed() to tell if
gossip is enabled or not , since timer set _armed = false before calling
the timer callback.
It was working correctly because we did not actually check is_enabled()
flag inside the timer callback but inside the send_gossip_digest_syn()'s
continuation and at that time the timer is armed again.
Use a standalone flag to do so.
Similar to a mutation_reader, but limited: it only returns whether a key
is sure not to exist in some mutation source. Non-blocking and expected
to execute fast. Corresponds to an sstable bloom filter.
To avoid ambiguity, it doesn't return a bool, instead a longer but less
ambiguous "definitely_doesnt_exists" or "maybe_exists".
* seastar 6f1dd3c...887f72d (8):
> finally(): don't discard any exception
> dpdk: check the resulting cluster for non-i40e NICs
> reactor: avoid SIGPIPE when writing to a socket
> memory: Don't run reclaimers if free memory is above the threshold
> core: Add missing include to transfer.hh
> dhcp: print the "sending discover" message only once
> reactor: count io_threaded_fallback statistic
> future: finally(): don't let the exceptional future to be ignored
"This series enables incremental eviction of data from cache. The eviction is
controlled by the LSA tracker, which consideres evictable regions as part of
its reclaim() method."
When LSA reclaimer cannot reclaim more space by compaction, it
will reclaim data by evicting from evictable regions.
Currently the only evictable region is the one owned by the row cache.
_lru_len may get stale when row_cache instance goes out of scope
purging all its partitions from cache. I'm assuming we're not really
interested in the number of partitions here, but rather a measure of
occupancy, so I applied a simple fix of using LSA region occupancy
instead.
Requiring alignment means that there must be 64K of contiguous space
to allocate each 32K segment. When memory is fragmented, we may fail
to allocate such segment, even though there's plenty of free space.
This especially hurts forward progress of compaction, which frees
segments randomly and relies on the fact that freeing a segment will
make it available to the next segment request.
Ok, shame on me: the version string was so obviously correct that I only
verified that the comparisons were working as expected.
Turns out it isn't: http://lists.boost.org/boost-users/2006/12/24194.php
boost::format will treat uint8_t arguments as char, and therefore we will end
up with the version string misprinted.
We can just cast it to uint16_t before we print, but since this is not exactly
a struct that we will be using all the time, let's favor readability over
saving a few bytes, and change all fields to uint16_t.
Signed-off-by: Glauber Costa <glommer@cloudius-systems.com>
We should have a better default here, that is prefixed by some directory. But
this will do for now, so we can start using the file without passing it as a
parameter all the time.
Also fix help string so it says scylla, and not cassandra.
Signed-off-by: Glauber Costa <glommer@cloudius-systems.com>
Aside from being the obviously correct thing to do, not having this will force us
to manually adjust num_tokens when running our sstables into Cassandra.
Signed-off-by: Glauber Costa <glommer@cloudius-systems.com>
It needs to access the non-existent "DatabaseDescriptor". Do as we have been doing,
and just pass the database object instead.
Signed-off-by: Glauber Costa <glommer@cloudius-systems.com>
Some code may attempt to use it during finalization after "instance"
was destroyed.
Reported by Pekka:
/usr/include/c++/4.9.2/bits/unique_ptr.h:291:14: runtime error:
reference binding to null pointer of type 'struct
standard_allocation_strategy'
./utils/allocation_strategy.hh:105:13: runtime error: reference
binding to null pointer of type 'struct standard_allocation_strategy'
./utils/allocation_strategy.hh:118:35: runtime error: reference
binding to null pointer of type 'struct allocation_strategy'
./utils/managed_bytes.hh:59:45: runtime error: member call on null
pointer of type 'struct allocation_strategy'
./utils/allocation_strategy.hh:82:9: runtime error: member access
within null pointer of type 'struct allocation_strategy'
As Avi suggested, we can use a tuple to make some comparisons more natural.
However, instead of doing a make_tuple on the comparison only, we can go
further and store the tuple internally.
I am still keeping the outer type, so it can host convenience functions like
to_sstring() and current().
Signed-off-by: Glauber Costa <glommer@cloudius-systems.com>
