In one of the following patches, we introduce support for zero-token
nodes. From that point, getting all nodes and getting all token
owners isn't equivalent. In this patch, we ensure that we consider
only token owners when we want to consider only token owners (for
example, in the replication logic), and we consider all nodes when
we want to consider all nodes (for example, in the topology logic).
The main purpose of this patch is to make the PR introducing
zero-token nodes easier to review. The patch that introduces
zero-token nodes is already complicated. We don't want trivial
changes from this patch to make noise there.
This patch introduces changes needed for zero-token nodes only in the
Raft-based topology and in the recovery mode. Zero-token nodes are
unsupported in the gossip-based topology outside recovery.
Some functions added to `token_metadata` and `topology` are
inefficient because they compute a new data structure in every call.
They are never called in the hot path, so it's not a serious problem.
Nevertheless, we should improve it somehow. Note that it's not
obvious how to do it because we don't want to make `token_metadata`
store topology-related data. Similarly, we don't want to make
`topology` store token-related data. We can think of an improvement
in a follow-up.
We don't remove unused `topology::get_datacenter_rack_nodes` and
`topology::get_datacenter_nodes`. These function can be useful in the
future. Also, `topology::_dc_nodes` is used internally in `topology`.
In one of the following patches, we introduce support for zero-token
nodes. A zero-token node that has successfully joined the cluster is
in the normal state but is not a normal token owner. Hence, the names
of `get_all_endpoints` and `get_all_ips` become misleading. They
should specify that the functions return only IDs/IPs of token owners.
Rather than allowing to keep both
host_id and endpoint, keep only one of them
and provide resolve functions that use the
token_metadata to resolve the host_id into
an inet_address or vice verse.
Signed-off-by: Benny Halevy <bhalevy@scylladb.com>
When topology barrier is blocked for longer than configured threshold
(2s), stale versions are marked as stalled and when they get released
they report backtrace to the logs. This should help to identify what
was holding for token metadata pointer for too long.
Example log:
token_metadata - topology version 30 held for 299.159 [s] past expiry, released at: 0x2397ae1 0x23a36b6 ...
Closesscylladb/scylladb#17427
range.hh was deprecated in bd794629f9 (2020) since its names
conflict with the C++ library concept of an iterator range. The name
::range also mapped to the dangerous wrapping_interval rather than
nonwrapping_interval.
Complete the deprecation by removing range.hh and replacing all the
aliases by the names they point to from the interval library. Note
this now exposes uses of wrapping intervals as they are now explicit.
The unit tests are renamed and range.hh is deleted.
Closesscylladb/scylladb#17428
This used to work before in replace-with-same-ip scenario, but
with host_id-s it's no longer relevant.
base_token_metadata has been removed from topology_change_info
because the conditions needed for its creation
are no longer met.
The token_metadata::get_normal_and_bootstrapping_token_to_endpoint_map
method was used only here. It's inlined in this
commit since it's too specific and incurs the overhead
of creating an intermediate map.
This commit fixes an inconsistency in method names:
get_host_id and get_host_id_if_known are
(internal_error, returns null), but there was only
one method for the opposite conversion - get_endpoint_for_host_id,
and it returns null. In this commit we change it to on_internal_error
if it can't find the argument and add another method
get_endpoint_for_host_id_if_known which returns null in this case.
We can't use get_endpoint_for_host_id/get_host_id
in host_id_or_endpoint::resolve since it's called
from storage_service::parse_node_list
-> token_metadata::parse_host_id_and_endpoint,
and exceptions are caught and handled in
`storage_service::parse_node_list`.
In this commit we enhance token_metadata with a pointer to the
new host_id-based generic_token_metadata specialisation (token_metadata2).
The idea is that in the following commits we'll go over all token_metadata
modifications and make the corresponding modifications to its new
host_id-based alternative.
The pointer to token_metadata2 is stored in the
generic_token_metadata::_new_value field. The pointer can be
mutable, immutable, or absent altogether (std::monostate).
It's mutable if this generic_token_metadata owns it, meaning
it was created using the generic_token_metadata(config cfg)
constructor. It's immutable if the
generic_token_metadata(lw_shared_ptr<const token_metadata2> new_value);
constructor was used. This means this old token_metadata is a wrapper for
new token_metadata and we can only use the get_new() method on it. The field
_new_value is empty for the new host_id-based token_metadata version.
The generic_token_metadata(std::unique_ptr<token_metadata_impl<NodeId>> impl, token_metadata2 new_value);
constructor is used for clone methods. We clone both versions,
and we need to pass a cloned token_metadata2 into constructor.
There are two overloads of get_new, for mutable and immutable
generic_token_metadata. Both of them throws an exception if
they can't get the appropriate pointer. There is also a
get_new_strong method, which returns an immutable owning
pointer. This is convenient since a lot of API's want an
owning pointer. We can't make the get_new/get_new_strong API
simpler and use get_new_strong everywhere since it mutate the
original generic_token_metadata by incrementing the reference
counter and this causes raises when it's passed between
shards in replicate_to_all_cores.
NodeId is used in all internal token_metadata data structures, that
previously used inet_address. We choose topology::key_kind based
on the value of the template parameter.
generic_token_metadata::update_topology overload with host_id
parameter is added to make update_topology_change_info work,
it now uses NodeId as a parameter type.
topology::remove_endpoint(host_id) is added to make
generic_token_metadata::remove_endpoint(NodeId) work.
pending_endpoints_for and endpoints_for_reading are just removed - they
are not used and not implemented. The declarations were left by mistake
from a refactoring in which these methods were moved to erm.
generic_token_metadata_base is extracted to contain declarations, common
to both token_metadata versions.
Templates are explicitly instantiated inside token_metadata.cc, since
implementation part is also a template and it's not exposed to the header.
There are no other behavioral changes in this commit, just syntax
fixes to make token_metadata a template.
In the next commits token_metadata will be
made a template with NodeId=inet_address|host_id
parameter. This parameter will be passed to dc_rack_fn
function, so it also should be made a template.
In subsequent commits we'll need the following api for token_metadata:
token_metadata(token_metadata2_ptr);
get_new() -> token_metadata2*
where token_metadata2 is the new version of token_metadata,
based on host_id.
In other words:
* token_metadata knows the new version of itself and returns a pointer
to it through get_new()
* token_metadata can be constructed based solely on the new version,
without its own implementation. In this case the only method we can
use on it is get_new.
This allows to pass token_metadata2 to API's with token_metadata in method
signature, if these APIs are known to only use the get_new method on the
passed token_metadata.
And back to topology_change_info - if we got it from the new token_metadata
we want to be able to construct token_metadata from token_metadata2 contained
in it, and this requires it to be a ptr, not value.
define token_metadata_ptr in token_metadata_fwd.hh
So that the declaration of `make_splitter` can be moved
to token_range_splitter.hh, where it belongs,
and so token_metadata.hh won't have to include it.
Signed-off-by: Benny Halevy <bhalevy@scylladb.com>
It's better to pass a disengaged optional when
the caller doesn't have the information rather than
passing the default dc_rack location so the latter
will never implicitly override a known endpoint dc/rack location.
Signed-off-by: Benny Halevy <bhalevy@scylladb.com>
Closes#15300
It's stored outside of topology table,
since it's updated not through RAFT, but
with a new 'fence' raft command.
The current value is cached in shared_token_metadata.
An initial fence version is loaded in main
during storage_service initialisation.
We use utils::phased_barrier. The new phase
is started each time the version is updated.
We track all instances of token_metadata,
when an instance is destroyed the
corresponding phased_barrier::operation is
released.
It's stored in as a static column in topology table,
will be updated at various steps of the topology
change state machine.
The initial value is 1, zero means that topology
versions are not yet supported, will be
used in RPC handling.
The function storage_service::update_pending_ranges is
turned to update_topology_changes_info.
The pending_endpoints and read_endpoints will be
computed later, when the erms are rebuilt.
We already use the new pending_endpoints from erm though
the get_pending_ranges virtual function, in this commit
we update all the remaining places to use the new
implementation in erm, as well as remove the old implementation
in token_metadata.
We plan to move pending_endpoints and read_endpoints, along
with their computation logic, from token_metadata to
vnode_effective_replication_map. The vnode_effective_replication_map
seems more appropriate for them since it contains functionally
similar _replication_map and we will be able to reuse
pending_endpoints/read_endpoints across keyspaces
sharing the same factory_key.
At present, pending_endpoints and read_endpoints are updated in the
update_pending_ranges function. The update logic comprises two
parts - preparing data common to all keyspaces/replication_strategies,
and calculating the migration_info for specific keyspaces. In this commit,
we introduce a new topology_change_info structure to hold the first
part's data add create an update_topology_change_info function to
update it. This structure will later be used in
vnode_effective_replication_map to compute pending_endpoints
and read_endpoints. This enables the reuse of topology_change_info
across all keyspaces, unlike the current update_pending_ranges
implementation, which is another benefit of this refactoring.
The update_topology_change_info implementation is mostly derived from
update_pending_ranges, there are a few differences though:
* replacing async and thread with plain co_awaits;
* adding a utils::clear_gently call for the previous value
to mitigate reactor stalls if target_token_metadata grows large;
* substituting immediately invoked lambdas with simple variables and
blocks to reduce noise, as lambdas would need to be converted into coroutines.
The original update_pending_ranges remains unchanged, and will be
removed entirely upon transitioning to the new implementation.
Meanwhile, we add an update_topology_change_info call to
storage_service::update_pending_ranges so that we can
iteratively switch the system to the new implementation.
In this patch we add
token_metadata::set_topology_transition_state method.
If the current state is
write_both_read_new update_pending_ranges
will compute new ranges for read requests. The default value
of topology_transition_state is null, meaning no read
ranges are computed. We will add the appropriate
set_topology_transition_state calls later.
Also, we add endpoints_for_reading method to get
read endpoints based on the computed ranges.
Now update_pending_ranges is quite complex, mainly
because it tries to act efficiently and update only
the affected intervals. However, it uses the function
abstract_replication_strategy::get_ranges, which calls
calculate_natural_endpoints for every token
in the ring anyway.
Our goal is to start reading from the new replicas for
ranges in write_both_read_new state. In the current
code structure this is quite difficult to do, so
in this commit we first simplify update_pending_ranges.
The main idea of the refactoring is to build a new version
of token_metadata based on all planned changes
(join, bootstrap, replace) and then for each token
range compare the result of calculate_natural_endpoints on
the old token_metadata and on the new one.
Those endpoints that are in the new version and
are not in the old version should be added to the pending_ranges.
The add_mapping function is extracted for the
future - we are going to use it to handle read mappings.
Special care is taken when replacing with the same IP.
The coordinator employs the
get_natural_endpoints_without_node_being_replaced function,
which excludes such endpoints from its result. If we compare
the new (merged) and current token_metadata configurations, such
endpoints will also be absent from pending_endpoints since
they exist in both. To address this, we copy the current
token_metadata and remove these endpoints prior to comparison.
This ensures that nodes being replaced are treated
like those being deleted.
Currently, scans are splitting partition ranges around tokens. This
will have to change with tablets, where we should split at tablet
boundaries.
This patch introduces token_range_splitter which abstracts this
task. It is provided by effective_replication_map implementation.
Don't maintain a "shadow" endpoint_to_host_id_map in token_metadata_impl.
Instead, get the nodes_by_endpoint map from topology
and use it to build the endpoint_to_host_id_map.
Signed-off-by: Benny Halevy <bhalevy@scylladb.com>
Add a simple node state model with:
`joining`, `normal`, `leaving`, and `left` states
to help managing nodes during replace
with the the same ip address.
Later on, this could also help prevent nodes
that were decommissioned, removed, or replaced
from rejoining the cluster.
Signed-off-by: Benny Halevy <bhalevy@scylladb.com>
And keep per node information (idx, host_id, endpoint, dc_rack, is_pending)
in node objects, indexed by topology on several indices like:
idx, host_id, endpoint, current/pending, per dc, per dc/rack.
The node index is a shorthand identifier for the node.
node* and index are valid while the respective topology instance is valid.
To be used, the caller must hold on to the topology / token_metadata object
(e.g. via a token_metadata_ptr or effective_replication_map)
Refs #6403
Signed-off-by: Benny Halevy <bhalevy@scylladb.com>
topology: add node idx
Signed-off-by: Benny Halevy <bhalevy@scylladb.com>
storage_service::replicate_to_all_cores has a sophisticated way
to mutate the token_metadata and effective_replication_map
on shard 0 and cloning those to all other shards, applying
the changes only mutate and clone succeeded on all shards
so we don't end up with only some of the shards with the mutated
copy if an error happend mid-way (and then we would need to
roll-back the change for exception safety).
shared_token_metadata::mutate_token_metadata is currently only called from
a unit test that needs to mutate the token metadata only on shard 0,
but a following patch will require doing that on all shards.
This change adds this capbility by enforcing the call to be
on shard 0m mutating the token_metdata into a temporary pending copy
and cloning it on all other shards. Only then, when all shard
succeeded, set the modified token_metadata on all shards.
Signed-off-by: Benny Halevy <bhalevy@scylladb.com>
Now, with a44ca06906,
is_normal_token_owner that replaced is_member
does not rely anymore on the pending status
of endpoints in topology.
With that we can get rid of this state and just keep
all endpoints we know about in the topology.
Signed-off-by: Benny Halevy <bhalevy@scylladb.com>
As described in #11993 per-shard repair_info instances get the effective_replication_map on their own with no centralized synchronization.
This series ensures that the effective replication maps used by repair (and other associated structures like the token metadata and topology) are all in sync with the one used to initiate the repair operation.
While at at, the series includes other cleanups in this area in repair and view that are not fixes as the calls happen in synchronous functions that do not yield.
Fixes#11993Closes#11994
* github.com:scylladb/scylladb:
repair: pass erm down to get_hosts_participating_in_repair and get_neighbors
repair: pass effective_replication_map down to repair_info
repair: coroutinize sync_data_using_repair
repair: futurize do_repair_start
effective_replication_map: add global_effective_replication_map
shared_token_metadata: get_lock is const
repair: sync_data_using_repair: require to run on shard 0
repair: require all node operations to be called on shard 0
repair: repair_info: keep effective_replication_map
repair: do_repair_start: use keyspace erm to get keyspace local ranges
repair: do_repair_start: use keyspace erm for get_primary_ranges
repair: do_repair_start: use keyspace erm for get_primary_ranges_within_dc
repair: do_repair_start: check_in_shutdown first
repair: get_db().local() where needed
repair: get topology from erm/token_metdata_ptr
view: get_view_natural_endpoint: get topology from erm
Topology is copied only from token_metadata_impl::clone_only_token_map
which copies the token_metadata_impl with yielding to prevent reactor
stalls. This should apply to topology as well, so
add a clone_gently function for cloning the topology
from token_metadata_impl::clone_only_token_map.
Signed-off-by: Benny Halevy <bhalevy@scylladb.com>
To be used for specifying nodes either by their
host_id or ip address and using the token_metadata
to resolve the mapping.
Signed-off-by: Benny Halevy <bhalevy@scylladb.com>
