Log

For each vnode in the ring, there exists one log file.
Currently the log is used for 3 things:
Loading the state of objects into the materialiser on restart (done by replaying the log).
Resending lost updates to external DCs.
Reading an older version of an object than what is available in memory.
Each log entry is defined by the following records:
-type log_operation() :: #log_operation{}.
-record(log_operation, {
    tx_id :: txid(),
    op_type :: update
             | prepare
             | commit
             | abort
             | noop,
    log_payload :: any_log_payload()
}).
​
​
-type any_log_payload() :: #update_log_payload{}
                         | #commit_log_payload{}
                         | #abort_log_payload{}
                         | #prepare_log_payload{}.
-type commit_log_payload() :: #commit_log_payload{}.
-type update_log_payload() :: #update_log_payload{}.
-type op_number() :: #op_number{}.
-type op_name() :: atom().
-type op_param() :: term().
-type op() :: {op_name(), op_param()}.
​
​
-record(commit_log_payload, {
    commit_time :: dc_and_commit_time(),
    snapshot_time :: snapshot_time()
}).
​
-record(update_log_payload, {
    key :: key(),
    bucket :: bucket(),
    type :: type(),
    op :: op()
}).
​
-record(abort_log_payload, {}).
​
-record(prepare_log_payload, {prepare_time :: non_neg_integer()}).
​
-record(op_number, {
    node :: undefined | {node(), dcid()},
    global :: undefined | non_neg_integer(),
    local :: undefined | non_neg_integer()
}).
Last modified 3yr ago