zrepl

zrepl

zrepl is a one-stop ZFS backup & replication solution.

This project is a fork of zrepl.

Current FreeBSD port is here. Keep in mind, sometimes it is a development version of this project, for testing.

Stable version of this project can be easy installed on FreeBSD using

pkg install zrepl-dsh2dsh

I don't build RPMs, but if somebody needs, here is a third party repo. Thanks to fluoros-jp for that.

Breaking changes!!!

• The project has switched from gRPC to REST API.

  This change isn't compatible with old job configurations. Instead of configuring serv: for every job, it configures in one place:

  # Include file with keys for accessing remote jobs and authenticate remote
  # clients. The filename is relative to filename of this configuration file.
  include_keys: "keys.yaml"
  # Include multiple files with keys from directory.
  #include_keys "keys.d/*.yaml"
  
  listen:
    # Serve "sink" and "source" jobs for network access.
    - addr: ":8888"
      tls_cert: "/usr/local/etc/ssl/cert.pem"
      tls_key: "/usr/local/etc/ssl/key.pem"
      zfs: true
  

  This configuration serves http and https API requests. tls_cert and tls_key are optional and needed for serving https requests.

  keys.yaml contains authentication keys of remote clients:

  # Clients with defined authentication keys have network access to "sink" and
  # "source" jobs. The key name is their client identity name.
  
  # Authentication token and client_identity for me.
  - name: "a.domain.com"          # client_identity
    key: "long and secret token"
  

  By default all authenticated clients have remote access to sink and source jobs. But it can be restricted using client_keys like:

  jobs:
    - name: "zdisk"
      type: "sink"
      # Restrict access to this job for listed remote clients
      client_keys:
        - "key1"
        - "key2"
        # and nobody else.
  

• All transports has been replaced by local and http transports.

  local transport configuration looks almost the same:

  jobs:
    - name: "zroot-to-zdisk"
      type: "push"
      connect:
        type: "local"
        listener_name: "zdisk"
        client_identity: "localhost"
  

  with one exception. listener_name now is a remote job name actually.

  The new http transport replaced all network transports. Its configuration look like:

  jobs:
    - name: "zroot-to-server"
      type: "push"
      connect:
        type: "http"
        server: "https://server:8888"
        listener_name: "zdisk"
        client_identity: "serverkey"
  
    - name: "server-to-zdisk"
      type: "pull"
      connect:
        type: "http"
        server: "https://server:8888"
        listener_name: "zroot-to-client"
        client_identity: "serverkey"
  

  listener_name is a job name on the server with type of sink or source.

  client_identity is a key name from keys.yaml. That key will be sent to the server for authentication and the server must have a key with the same key content in keys.yaml. name can be different, because sink and source jobs use key name as client_identity.

Changes from upstream:

• Fresh dependencies

• last_n keep rule fixed Merged.

  See #691.

• Some of resolved upstream issues:

  • zrepl/zrepl#86
  • zrepl/zrepl#170
  • zrepl/zrepl#287
  • zrepl/zrepl#324
  • zrepl/zrepl#394
  • zrepl/zrepl#422
  • zrepl/zrepl#498
  • zrepl/zrepl#507
  • zrepl/zrepl#708
  • zrepl/zrepl#717
  • zrepl/zrepl#745
  • zrepl/zrepl#748

• New dataset filter syntax instead of filesystems:

  New field datasets is a list of patterns. By default a pattern includes matched dataset. All patterns applied in order and last matched pattern wins. Lets see some examples.

  The following configuration will allow access to all datasets:

  jobs:
    - name: "source"
      type: "source"
      datasets:
        - recursive: true
  

  The following configuration will allow access to datasets zroot/ROOT/default and zroot/usr/home including all their children.

  jobs:
    - name: "snap-1h"
      type: "snap"
      datasets:
        - pattern: "zroot/ROOT/default"
          recursive: true
        - pattern: "zroot/usr/home"
          recursive: true
  

  The following configuration is more complicated:

  jobs:
    - name: "source"
      type: "source"
      datasets:
        - pattern: "tank"         # rule (1)
          recursive: true
        - pattern: "tank/foo"     # rule (2)
          exclude: true
          recursive: true
        - pattern: "tank/foo/bar" # rule (3)
  

  tank/foo/bar/loo is excluded by (2), because (3) isn't matched (it isn't recursive).

  tank/bar is included by (1).

  tank/foo/bar is included by (3), because yes, it matched by (2), but last matched rule wins and (3) is the last matched rule.

  zroot isn't included at all, because nothing matched it.

  tank/var/log is included by (1), becuase this rule is recursive and other rules are not matched.

  For compatibility reasons old filesystems still works, but I wouldn't suggest use it. It's deprecated and can be removed anytime.

• Added support of shell patterns for datasets definitions.

  Configuration example:

  datasets:
    # exclude all children of zroot/bastille/jails
    - pattern: "zroot/bastille/jails"
      exclude: true
      recursive: true
    # except datasets matched by this shell pattern
    - pattern: "zroot/bastille/jails/*/root"
      shell: true
  

  This configuration includes zroot/bastille/jails/a/root, zroot/bastille/jails/b/root zfs datasets, and excludes zroot/bastille/jails/a, zroot/bastille/jails/b zfs datasets on.

  Another example:

  datasets:
    # exclude datasets matched by this shell pattern
    - pattern: "zroot/bastille/jails/*/root"
      exclude: true
      shell: true
    # and include everything else inside zroot/bastille/jails
    - pattern: "zroot/bastille/jails"
      recursive: true
  

  excludes zroot/bastille/jails/a/root, zroot/bastille/jails/b/root and includes everything else inside zroot/bastille/jails.

  See Match for details about patterns.

• Added new log formatters: json and text.

  Both formatters use slog for formatting log entries. The new json formatter replaces old json formatter. Configuration example:

  logging:
    - type: "file"
      format: "text"            # or "json"
      time: false               # don't prepend with date and time
      hide_fields:
        - "span"                # don't log "span" field
  

• Added ability to log into a file.

  See #756. Configuration example:

  logging:
    - type: "file"
      format: "text"            # or "json"
      time: false               # don't prepend with date and time
      hide_fields: &hide-log-fields
        - "span"                # don't log "span" field
      level:  "error"           # log errors only
      # without filename logs to stderr
    - type: "file"
      format: "text"
      hide_fields: *hide-log-fields
      level:  "info"
      filename: "/var/log/zrepl.log"
  

• Replication jobs (without periodic snapshotting) can be configured for periodic run.

  See #758. Configuration example:

  - name: "zroot-to-server"
    type: "push"
    interval: "1h"
    snapshotting:
      type: "manual"
  

  Both pull and push job types support configuration of periodic run using cron specification. For instance:

  - name: "zroot-to-server"
    type: "push"
    cron: "25 15-22 * * *"
    snapshotting:
      type: "manual"
  

  See CRON Expression Format for details.

• Added ability to configure command pipelines between zfs send and zfs recv.

  See #761. Configuration example:

  send:
    execpipe:
      # zfs send | zstd | mbuffer
      - [ "zstd", "-3" ]
      - [ "/usr/local/bin/mbuffer", "-q", "-s", "128k", "-m", "100M" ]
  

  recv:
    execpipe:
      # mbuffer | unzstd | zfs receive
      - [ "/usr/local/bin/mbuffer", "-q", "-s", "128k", "-m", "100M" ]
      - [ "unzstd" ]
  

  zrepl exports some zfs send|recv args as env variables: ZREPL_SEND_RESUME_TOKEN, ZREPL_SEND_FROM, ZREPL_SEND_SNAPSHOT and ZREPL_RECV_FS.

• Added Icinga/Nagios checks for checking the daemon is alive, snapshots count is ok, latest or oldest snapshots are not too old.

  See #765. Configuration example:

  monitor:
    count:
      - prefix: "zrepl_frequently_"
        warning: 20
        critical: 30
      - prefix: "zrepl_hourly_"
        warning: 31
        critical: 50
      - prefix: "zrepl_daily_"
        warning: 91
        critical: 92
      - prefix: "zrepl_monthly_"
        warning: 13
        critical: 14
      - prefix: ""            # everything else
        warning: 2
        critical: 10
    latest:
      - prefix: "zrepl_frequently_"
        critical: "48h"       # 2d
      - prefix: "zrepl_hourly_"
        critical: "48h"
      - prefix: "zrepl_daily_"
        critical: "48h"
      - prefix: "zrepl_monthly_"
        critical: "768h"      # 32d
    oldest:
      - prefix: "zrepl_frequently_"
        critical: "48h"       # 2d
      - prefix: "zrepl_hourly_"
        critical: "168h"      # 7d
      - prefix: "zrepl_daily_"
        critical: "2208h"     # 90d + 2d
      - prefix: "zrepl_monthly_"
        critical: "8688h"     # 30 * 12 = 360d + 2d
      - prefix: ""            # everything else
        critical: "168h"      # 7d
  

  Every item can be configured to skip some datasets from the check, like:

  - prefix: "zrepl_monthly_"
    skip_datasets:
      - pattern: "zdisk/video"
    warning: 13
    critical: 14
  

  In this example it checks number of snapshots with prefix zrepl_monthly_ for every dataset, configured in datasets, except zdisk/video. skip_datasets has the same syntax, like datasets.

  An example of a daily script:

  echo
  echo "zrepl status:"
  zrepl monitor alive
  zrepl monitor snapshots
  

• Removed support of postgres-checkpoint and mysql-lock-tables hooks.

• Periodic snapshotting now recognizes cron specification. For instance:

  snapshotting:
    type: "periodic"
    cron: "25 15-22 * * *"
  

  type: "cron" still works too, just for compatibility. Both of them is the same type.

• Fast skip "keep all" pruning.

  Instead of configuration like this:

  pruning:
    keep:
      - type: "regex"
        regex: ".*"
  

  or like this:

  pruning:
    keep_sender:
      - type: "regex"
        regex: ".*"
    keep_receiver:
  

  which keeps all snapshots, now it's possible to omit pruning: at all, or just one of keep_sender: or keep_receiver:. In this case zrepl will early abort pruning and mark it as done.

  Originally zrepl requests all snapshots and does nothing after that, because pruning configured to keep all snapshots, but anyway it spends some time executing zfs commands.

• Snapshots are named using local time for timestamps, instead of UTC.

  So instead of snapshot names like zrepl_20240508_140000_000 it's zrepl_20240508_160000_CEST. timestamp_local defines time zone of timestamps. By default it's local time, but with timestamp_local: false it's UTC. Configuration like:

  snapshotting:
    type: "periodic"
    cron: "*/15 * * * *"
    prefix: "zrepl_"
    timestamp_format: "20060102_150405_000"
    timestamp_local: false
  

  returns original naming like zrepl_20240508_140000_000 with UTC time.

• Configurable RPC timeout (1 minute by default). Configuration example:

  global:
    rpc_timeout: "2m30s"
  

  sets RPC timeout to 2 minutes and 30 seconds.

  See also zrepl/zrepl#791

• Configurable path to zfs binary ("zfs" by default). Configuration example:

  global:
    zfs_bin: "/sbin/zfs"
  

  sets zfs binary path to "/sbin/zfs".

• Replication now generates a stream package that sends all intermediary snapshots (zfs send -I), instead of every intermediary snapshot one by one (zfs send -i). Such replication is much faster. For instance a replication job on my desktop configured like:

  replication:
    concurrency:
      steps: 4
      size_estimates: 8
  

  replicates over WLAN for 1m32s, instead of 8m.

• New command zrepl signal stop

  Stop the daemon right now. Actually it's the same like sending SIGINT to the daemon.

• New command zrepl signal shutdown

  Stop the daemon gracefully. After this signal, zrepl daemon will exit as soon as it'll be safe. It interrupts any operation, except replication steps. The daemon will wait for all replication steps completed and exit.

  Sending SIGTERM has the same effect.

• Redesigned zrepl status

• zfs send -w is default now. Example how to change it back:

  send:
    raw: false
  

• New configuration for control and prometheus services. Example:

  listen:
    # control socket for zrepl client, like `zrepl signal` or `zrepl status`.
    - unix: "/var/run/zrepl/control"
      # unix_mode: 0o660            # write perm for group
      control: true
  
    # Export Prometheus metrics on http://127.0.0.1:8000/metrics
    - addr: "127.0.0.1:8000"
      # tls_cert: "/usr/local/etc/zrepl/cert.pem"
      # tls_key: "/usr/local/etc/zrepl/key.pem"
      metrics: true
  

  One of addr or unix is required or both of them can be configured. One of control or metrics is required or both of them can be configured too. Everything else is optional. For backward compatibility old style configuration works too.

  See also zrepl/zrepl#780

• New optional pre and post hooks for push and pull jobs. Example:

  - name: "zroot-to-zdisk"
    type: "push"
    hooks:
      pre:
        path: "/root/bin/zrepl_hook.sh"
        args: [ "pre" ]         # optional positional parameters
        env:                    # optional environment variables
          ZREPL_FOOBAR: "foo"
        timeout: "30s"          # optional, default is 1m
        # don't continue job if exit status is nonzero (default: false)
        err_is_fatal: true
      post:
        path: "/root/bin/zrepl_hook.sh"
        args: [ "post" ]        # optional positional parameters
        env:                    # optional environment variables
          ZREPL_FOOBAR: "bar"
        timeout: "0s"           # without timeout at all
  

  This configuration runs /root/bin/zrepl_hook.sh pre before replication with environment variables:

  ZREPL_FOOBAR=foo
  ZREPL_JOB_NAME=zroot-to-zdisk
  

  If it exit with nonzero exit status the job will not continue. By default err_is_fatal: false and exit status is ignored.

  After pruning finished it runs /root/bin/zrepl_hook.sh post with environment variables:

  ZREPL_FOOBAR=bar
  ZREPL_JOB_ERR=
  ZREPL_JOB_NAME=zroot-to-zdisk
  

  The post hook sets ZREPL_JOB_ERR to the last error. It's empty if the job finished without errors.

• The pruning now prunes filesystems concurrently.

  By default it uses the number of CPUs as concurrency limit and it can be changed in config:

  jobs:
    - name: "zroot-to-zdisk"
      pruning:
        concurrency: 1
  

  That concurrency can be defined on both sides: local and remote. One side uses it to limit concurrent zfs list and other side uses concurrency from their config to limit concurrent zfs destroy.

• Job configurations can be included from multiple files:

  include_jobs: "jobs.d/*.yaml"
  

  Like include_keys, the directory is relative to main configuration file.

  include_jobs can be combined with jobs:

  jobs:
    - name: "zroot-to-zdisk"
  include_jobs: "jobs.d/*.yaml"
  

• The replication can be configured to replicate subset of snapshots, instead of all of them. Example:

  jobs:
    - name: "zroot-to-zdisk"
      type: "push"
      replication:
        prefix: "zrepl_"
  

  This configuration will replicate snapshots with names beginning with "zrepl_".

  See also zrepl/zrepl#403

• Snapshots now created concurrently.

  By default it uses the number of CPUs as concurrency limit and it can be changed in config:

  jobs:
    - name: "zroot-to-zdisk"
      snapshotting:
        type: "periodic"
        concurrency: 1
  

• Recursive snapshots zfs snaphot -r

  Where possible, snapshots created using zfs snaphot -r. With configuration like this:

  jobs:
    - name: "snap-1h"
      type: "snap"
      datasets:
        - pattern: "zroot"
          recursive: true
  

  zfs snapshot -r zroot recursively create snapshots of all descendent datasets, instead of one by one. But if recursive dataset has any exclusion, like this:

  jobs:
    - name: "snap-1h"
      type: "snap"
      datasets:
        - pattern: "zroot"
          recursive: true
        - pattern: "zroot/foo"
          exclude: true
  

  -r not used and it create snapshots of all descendent datasets one by one.

  See also zrepl/zrepl#634

• Faster zfs list (zrepl/zrepl#870)

  Runs zfs list filesystem..., if possible, and list only specific datasets, instead of zfs list, which lists every dataset on the entire system.

• Don't create empty or small snapshots

  New option written_threshold defines the amount of space that should be written since the previous snapshot, before the new snapshot can be created.

  With configuration like this:

  snapshotting:
    type: "periodic"
    written_threshold: 1024
  

  any dataset, which has written less than 1024 bytes, will be skipped.

  See also zrepl/zrepl#728

• Manual "periodic" snapshot jobs

  Snapshot jobs can be configured don't run periodically, but on signal only, from CLI or zrepl status, like this:

  snapshotting:
    type: "periodic"
    interval: "manual"
  

Upstream user documentation

User Documentation can be found at zrepl.github.io. Keep in mind, it doesn't contain changes from this fork.