3.14. Replicator

3.14.1. Replicator Database Configuration

[replicator]

max_jobs

Added in version 2.1.

Number of actively running replications. This value represents the threshold to trigger the automatic replication scheduler. The system will check every interval milliseconds how many replication jobs are running, and if there are more than max_jobs active jobs, the scheduler will pause-and-restart up to max_churn jobs in the scheduler queue. Making this value too high could cause performance issues, while making it too low could mean replications jobs might not have enough time to make progress before getting unscheduled again. This parameter can be adjusted at runtime and will take effect during next rescheduling cycle:

[replicator]
max_jobs = 500

interval

Added in version 2.1.

Scheduling interval in milliseconds. During each reschedule cycle the scheduler might start or stop up to max_churn number of jobs:

[replicator]
interval = 60000

max_churn

Added in version 2.1.

Maximum number of replication jobs to start and stop during rescheduling. This parameter, along with interval, defines the rate of job replacement. During startup, however, a much larger number of jobs could be started (up to max_jobs) in a short period of time:

[replicator]
max_churn = 20

max_history

Maximum number of events recorded for each job. This parameter defines an upper bound on the consecutive failure count for a job, and in turn the maximum backoff factor used when determining the delay before the job is restarted. The longer the length of the crash count, the longer the possible length of the delay:

[replicator]
max_history = 20

update_docs

Added in version 2.1.

When set to true replicator will update replication document with error and triggered states. This approximates pre-2.1 replicator behavior:

[replicator]
update_docs = false

worker_batch_size

With lower batch sizes checkpoints are done more frequently. Lower batch sizes also reduce the total amount of used RAM memory:

[replicator]
worker_batch_size = 500

worker_processes

More worker processes can give higher network throughput but can also imply more disk and network IO:

[replicator]
worker_processes = 4

http_connections

Maximum number of HTTP connections per replication:

[replicator]
http_connections = 20

connection_timeout

HTTP connection timeout per replication. This is divided by three (3) when the replicator makes changes feed requests. Even for very fast/reliable networks it might need to be increased if a remote database is too busy:

[replicator]
connection_timeout = 30000

retries_per_request

Changed in version 2.1.1.

If a request fails, the replicator will retry it up to N times. The default value for N is 5 (before version 2.1.1 it was 10). The requests are retried with a doubling exponential backoff starting at 0.25 seconds. So by default requests would be retried in 0.25, 0.5, 1, 2, 4 second intervals. When number of retires is exhausted, the whole replication job is stopped and will retry again later:

[replicator]
retries_per_request = 5

socket_options

Some socket options that might boost performance in some scenarios:

{nodelay, boolean()}
{sndbuf, integer()}
{recbuf, integer()}
{priority, integer()}

See the inet Erlang module’s man page for the full list of options:

[replicator]
socket_options = [{keepalive, true}, {nodelay, false}]

valid_socket_options

Added in version 3.3.

Valid socket options. Options not in this list are ignored. Most of those options are low level and setting some of them may lead to unintended or unpredictable behavior. See inet Erlang docs for the full list of options:

[replicator]
valid_socket_options = buffer,keepalive,nodelay,priority,recbuf,sndbuf

ibrowse_options

Added in version 3.4: A non-default ibrowse setting is needed to support IPV6-only replication sources or targets:

{prefer_ipv6, boolean()}

See the ibrowse site for the full list of options:

[replicator]
ibrowse_options = [{prefer_ipv6, true}]

valid_ibrowse_options

Added in version 3.4.

Valid ibrowse options. Options not in this list are ignored:

[replicator]
valid_ibrowse_options = prefer_ipv6

valid_endpoint_protocols

Added in version 3.3.

Valid replication endpoint protocols. Replication jobs with endpoint urls not in this list will fail to run:

[replicator]
valid_endpoint_protocols = http,https

valid_endpoint_protocols_log

Added in version 3.4.

When enabled, CouchDB will log any replication that uses the insecure http protocol:

[replicator]
valid_endpoint_protocols_log = true

verify_ssl_certificates_log

Added in version 3.4.

When enabled, and if ssl_trusted_certificates_file is configured but verify_ssl_certificates is not, CouchDB will check the validity of the TLS certificates of all sources and targets ( without causing the replication to fail) and log any issues:

[replicator]
verify_ssl_certificates_log = true

valid_proxy_protocols

Added in version 3.3.

Valid replication proxy protocols. Replication jobs with proxy urls not in this list will fail to run:

[replicator]
valid_proxy_protocols = http,https,socks5

checkpoint_interval

Added in version 1.6.

Defines replication checkpoint interval in milliseconds. Replicator will requests from the Source database at the specified interval:

[replicator]
checkpoint_interval = 5000

Lower intervals may be useful for frequently changing data, while higher values will lower bandwidth and make fewer requests for infrequently updated databases.

use_checkpoints

Added in version 1.6.

If use_checkpoints is set to true, CouchDB will make checkpoints during replication and at the completion of replication. CouchDB can efficiently resume replication from any of these checkpoints:

[replicator]
use_checkpoints = true

Note

Checkpoints are stored in local documents on both the source and target databases (which requires write access).

Warning

Disabling checkpoints is not recommended as CouchDB will scan the Source database’s changes feed from the beginning.

use_bulk_get: Added in version 3.3.

If use_bulk_get is true, CouchDB will attempt to use the _bulk_get HTTP API endpoint to fetch documents from the source. Replicator should automatically fall back to individual doc GETs on on error; however, in some cases it may be useful to prevent spending time attempting to call _bulk_get altogether.

cert_file

Path to a file containing the user’s certificate:

[replicator]
cert_file = /full/path/to/server_cert.pem

key_file

Path to file containing user’s private PEM encoded key:

[replicator]
key_file = /full/path/to/server_key.pem

password

String containing the user’s password. Only used if the private key file is password protected:

[replicator]
password = somepassword

verify_ssl_certificates

Set to true to validate peer certificates:

[replicator]
verify_ssl_certificates = false

ssl_trusted_certificates_file

File containing a list of peer trusted certificates (in the PEM format):

[replicator]
ssl_trusted_certificates_file = /etc/ssl/certs/ca-certificates.crt

ssl_certificate_max_depth

Maximum peer certificate depth (must be set even if certificate validation is off):

[replicator]
ssl_certificate_max_depth = 3

auth_plugins

Added in version 2.2.

List of replicator client authentication plugins. Plugins will be tried in order and the first to initialize successfully will be used. By default there are two plugins available: couch_replicator_auth_session implementing session (cookie) authentication, and couch_replicator_auth_noop implementing basic authentication. For backwards compatibility, the no-op plugin should be used at the end of the plugin list:

[replicator]
auth_plugins = couch_replicator_auth_session,couch_replicator_auth_noop

usage_coeff

Added in version 3.2.0.

Usage coefficient decays historic fair share usage every scheduling cycle. The value must be between 0.0 and 1.0. Lower values will ensure historic usage decays quicker and higher values means it will be remembered longer:

[replicator]
usage_coeff = 0.5

priority_coeff

Added in version 3.2.0.

Priority coefficient decays all the job priorities such that they slowly drift towards the front of the run queue. This coefficient defines a maximum time window over which this algorithm would operate. For example, if this value is too small (0.1), after a few cycles quite a few jobs would end up at priority 0, and would render this algorithm useless. The default value of 0.98 is picked such that if a job ran for one scheduler cycle, then didn’t get to run for 7 hours, it would still have priority > 0. 7 hours was picked as it was close enough to 8 hours which is the default maximum error backoff interval:

[replicator]
priority_coeff = 0.98

3.14. Replicator

3.14.1. Replicator Database Configuration

3.14.2. Fair Share Replicator Share Allocation