This update for galera-4 fixes the following issues:

- libboost_system was removed upstream, fixes boo#1249900

Update to 26.4.22:

  * New feature for applications to define sequential consistency parameters for
    the certification process, via wsrep_certify_v1() which takes an application
  defined callback
  * Fixed assertions in report_last_committed() that could cause issues under
  certain thread scheduling conditions
  * Removed unnecessary assertions that could trigger due to normal timing
  variations in distributed environments
  * Fixed string format warnings that were appearing during builds on Debian Sid

26.4.21 changelog:

  * remove the dependency on openssl/engine.h - fixes codership/galera#663
  * writeset serialization during IST fails because the seqno locking mechanism
  does not prevent underlying data stores from discarding released actions,
  requiring the locked seqno to be passed to stores to enforce discard limits.
  * fixed ssl_cipher parameter type to string
  * when a node progresses from JOINED to SYNCED, a race condition may prevent it
  from being counted in the commit cut; ensure it is included upon announcing
  SYNCED and delay advancing the commit cut until the node's last applied value
  exceeds the current commit cut. GCS protocol version bumped.
  * expose versions of all protocols agreed on by quorum via status call.

26.4.20 changelog:

  * improve error and warning messages handling in the library. System error
  numbers and messages are hidden from thrown exceptions, except for the cases
  where the exception is thrown because of error from system call. Some warning
  level messages in gcomm and GCS have been changed to info level in cases where
  the message does not require action from the administrator.

Update to 26.4.19:

  * WSREP_KEY_SHARED key type is being phased out since it had insufficiently
    strict certification rules that may lead to certification test passes where it
    should fail. WSREP_KEY_REFERENCE fixes this, and all new writesets will be
    constructed using this. WSREP_KEY_SHARED is supported for backwards
    compatibility.
  * A memory leak in the commit cut action has been fixed.
  * Network level node isolation, which can help shut down all communication with
    the rest of the cluster to achieve a fail-silent crash. 
  * GCS level protocol bumped, as error voting for joiner in the JOINED state was
    broken when group-wide commit cut (implied SUCCESS vote) was not taken into
    account when processing error vote requests from the JOINED node. 
- Switch to mariadb fork of galera-4.

update to 26.4.18:

  * The garbd process hangs due to exceptions in the GCS layer not
    being caught within the receive loop, necessitating a forceful
    termination. Fixing this requires catching all exceptions
    within the loop, closing the connection first, and then
    continuing until a 'SELF-LEAVE' message ensures graceful exit.
  * When using SSL, crashes of garbd during graceful node shutdowns
    may occur, and result in the cluster becoming non-Primary;
    garbd now ignores the SIGPIPE signal to prevent this.
  * socket_ssl_compression is deprecated, now it isn't attempt to
    be set, and the user receives a warning if it was explicitly
    set.
  * Fixed commit cut tracking on node leave, and bumped GCS protocol
    version for backwards compatibility.
  * Executing the gcomm join process in the database server thread
    can lead to issues during allowlist callbacks, which should
    only be handled by Galera service threads. The gcomm join
    process is run within the gcomm service thread now.

Update to 26.4.17:

  - "Zero-level" certification keys are implemented. Currently there are 3 levels
  of resource keys: schema, table, row, yet one common resource used by every
  writeset, a node. The key is added by default to every writeset with the default
  type WSREP_KEY_SHARED. Now, node-wide operations can be performed with the same
  semantics as regular transactions.
  - Non-SYNCED nodes are not counted when calculating commit cut - maximum seqno
  that is no longer needed in certification index; situations may arise when group
  commit cut is applied before all preceding events are committed on a non-SYNCED
  node, now drain all monitors to ensure that preceding events are committed.
  - Stability of node joining process for gmcast improved when new nodes are
  connecting concurrently as nodes previously may arrive at a different conclusion
  on which nodes will be blocked, however now, connections are explicitly allowed
  from all nodes which may be accepted as members in the next view.
  - Stability of node joining process for EVS improved by dropping EVS join
  messages without own identified in joining state, to avoid starting membership
  protocol before at least one of the existing nodes has seen at least one join
  message from a joiner.
  - Improve "Failed to report last committed" warning to make it more accurate and
  user-friendly.

Update to 26.4.16:

  - Removed out-of-order certification for BF aborted transactions.
  - Better error handling when the socket is closed, also cleaned up excessive
  debug logging.

Update to 26.4.15:

  - an assertion is relaxed when Galera is configured with commit order disabled
  - in certificate index preload, an assertion in handle_ist_trx_preload() fails
  if the first handled IST preload write set has failed certification, and
  incrementing the index position caused the certificate index to end up with the
  wrong position; now fixed by appending dummy writeset sequence numbers into
  Certification::trx_map_
  - ASAN reported memory leak if gcs is closed after becoming joined but before
  reaching synced