engine-params.org

EP Parameters

1 Configuring the Engine

Engine parameters may be specified using the -e option to memcached. These parameters control various aspects of the behavior of the engine.

For example, if you would like to store your database in /data/base and prevent it from having objects larger than 1MB, you would run memcached like this:

memcached -E ep.so -e 'dbfile=/data/base;max_item_size=1048576'

2 Parameters for the EP Engine

key	type	descr
config_file	string	Path to additional parameters.
dbname	string	Path to on-disk storage.
shardpattern	string	File pattern for shards (see below)
ht_locks	int	Number of locks per hash table.
ht_size	int	Number of buckets per hash table.
initfile	string	Optional SQL script to run after
		opening DB
postInitfile	string	Optional SQL script to run after
		all DB shards and statements have
		been initialized
max_item_size	int	Maximum number of bytes allowed for
		an item.
max_size	int	Max cumulative item size in bytes.
max_txn_size	int	Max number of disk mutations per
		transaction.
mem_high_wat	int	Automatically evict when exceeding
		this size.
mem_low_wat	int	Low water mark to aim for when evicting.
min_data_age	int	Minimum data stability time before
		persist.
queue_age_cap	int	Maximum queue time before forcing persist.
tap_backlog_limit	int	Max number of items allowed in a
		tap backfill
tap_noop_interval	int	Number of seconds between a noop is sent
		on an idle connection
tap_keepalive	int	Seconds to hold open named tap connections
tap_bg_max_pending	int	Maximum number of pending bg fetch
		operations
		a tap queue may issue (before it must wait
		for responses to appear.
tap_backoff_period	float	Number of seconds the tap connection
		should back off after receiving ETMPFAIL
vb0	bool	If true, start with an active vbucket 0
waitforwarmup	bool	Whether to block server start during
		warmup.
warmup	bool	Whether to load existing data at startup.
expiry_window	int	expiry window to not persist an object
		that is expired (or will be soon)
exp_pager_stime	int	Sleep time for the pager that purges
		expired objects from memory and disk
failpartialwarmup	bool	If false, continue running after failing
		to load some records.
max_vbuckets	int	Maximum number of vbuckets expected (1024)
db_shards	int	Number of shards for db store
db_strategy	string	DB store strategy (“multiDB”, “singleDB”
		or “singleMTDB”)
vb_del_chunk_size	int	Chunk size of vbucket deletion
vb_chunk_del_time	int	vb chunk deletion threshold time (ms) used
		for adjusting the chunk size dynamically
concurrentDB	bool	True (default) if concurrent DB reads are
		permitted where possible.
chk_remover_stime	int	Interval for the checkpoint remover that
		purges closed unreferenced checkpoints.
chk_max_items	int	Number of max items allowed in a
		checkpoint
chk_period	int	Time bound (in sec.) on a checkpoint
max_checkpoints	int	Number of max checkpoints allowed per
		vbucket
inconsistent_slave_chk	bool	True if we allow a “downstream” master to
		receive checkpoint begin/end messages
item_num_based_new_chk	bool	Enable a new checkpoint creation if the
		number of items in a checkpoint is greater
		than the max number allowed
		along with normal get/set operations.
tap_backfill_resident	float	Resident item threshold for only memory
		backfill to be kicked off
keep_closed_chks	bool	True if we want to keep closed checkpoints
		in memory if the current memory usage is
		below high water mark
chk_meta_items_only	bool	True if a checkpoint should contain items
		with keys and meta data only.
bf_resident_threshold	float	Resident item threshold for only memory
		backfill to be kicked off
getl_default_timeout	int	The default timeout for a getl lock in (s)
getl_max_timeout	int	The maximum timeout for a getl lock in (s)
mutation_mem_threshold	float	Memory threshold on the current bucket
		quota for accepting a new mutation
tap_throttle_queue_cap	int	The maximum size of the disk write queue
		to throttle down tap-based replication. -1
		means don’t throttle.
tap_throttle_threshold	float	Percentage of memory in use before we
		throttle tap streams
klog_path	string	Path to the mutation key log.
klog_block_size	int	Mutation key log block size.
klog_flush	string	When to force buffer flushes during
		klog (off, commit1, commit2, full)
klog_sync	string	When to fsync during klog.

2.1 Shard Patterns

When using a multi-sharded database, shardpattern specifies how shards are named relative to the main database.

shardpattern uses a printf-style string to expand the path with the given variables:

• %d : The directory name containing the main database (This will be . for relative paths).
• %b : The base name of the main database.
• %i : The shard number.

The default value of shardpattern is %d/%b-%i.sqlite