tiflash.yaml

# 0. The tiflash image needs to be pushed to a Docker registry, and the following `image` field need to be adjusted to the registry address.
# 1. helm install pingcap/tidb-cluster --name=${TIDB_CLUSTER_NAME} --namespace=${MY-NAMESPACE} --version=v1.0.2 --set pd.image=tiflash,tikv.image=tiflash,tidb.image=tiflash
# 2. Replace the following TIDB_CLUSTER_NAME with the real tidb cluster name.
# 3. kubectl apply -f tiflash.yaml
apiVersion: apps/v1
kind: StatefulSet
metadata:
  name: rngine
spec:
  selector:
    matchLabels:
      app: rngine
  serviceName: rngine
  replicas: 1
  template:
    metadata:
      labels:
        app: rngine
    spec:
      containers:
      - name: rngine
        image: tiflash
        command:
        - /tikv-server-rngine
        - --addr=0.0.0.0:20160
        - --advertise-addr=${MY_POD_NAME}.rngine.${MY_NAMESPACE}.svc:20160
        - --data-dir=/data/rngine
        - --pd=TIDB_CLUSTER_NAME-pd.${MY_NAMESPACE}.svc:2379
        - --config=/etc/rngine/rngine.toml
        env:
        - name: MY_POD_NAME
          valueFrom:
            fieldRef:
              fieldPath: metadata.name
        - name: MY_NAMESPACE
          valueFrom:
            fieldRef:
              fieldPath: metadata.namespace
        volumeMounts:
        - name: data
          mountPath: /data
        - name: config
          mountPath: /etc/rngine
      volumes:
      - name: config
        configMap:
          name: tiflash
          items:
          - key: rngine.toml
            path: rngine.toml
  volumeClaimTemplates:
  - metadata:
      name: data
    spec:
      accessModes: [ "ReadWriteOnce" ]
      storageClassName: "local-storage"
      resources:
        requests:
          storage: 10Gi
---
apiVersion: apps/v1
kind: StatefulSet
metadata:
  name: tiflash
spec:
  selector:
    matchLabels:
      app: tiflash
  serviceName: tiflash
  replicas: 1
  template:
    metadata:
      labels:
        app: tiflash
    spec:
      containers:
      - name: tiflash
        image: tiflash
        command:
        - /tiflash
        - server
        - --config-file=/etc/tiflash/config.xml
        volumeMounts:
        - name: data
          mountPath: /data
        - name: config
          mountPath: /etc/tiflash
        - name: logs
          mountPath: /logs
      # TODO: decide if flash-cluster-manager should be deployed along side with tiflash
      - name: flash-cluster-manager
        image: tiflash
        command:
        - /flash-cluster-manager
        - /etc/tiflash/config.xml
        volumeMounts:
        - name: data
          mountPath: /data
        - name: config
          mountPath: /etc/tiflash
      # To better support k8s, allow normal log to output stdout,
      # so only the error log tailer is needed
      - name: tiflash-log
        image: busybox
        command:
        - /bin/sh
        - -c
        - |
          touch /logs/tiflash-server.log
          tail -n0 -F /logs/tiflash-server.log
        volumeMounts:
        - name: logs
          mountPath: /logs
      - name: error-log
        image: busybox
        command:
        - /bin/sh
        - -c
        - |
          touch /logs/tiflash-error.log
          tail -n0 -F /logs/tiflash-error.log
        volumeMounts:
        - name: logs
          mountPath: /logs
      - name: flash-cluster-manager-log
        image: busybox
        # TODO: The log directory is weird, can we just log to stdout?
        command:
        - /bin/sh
        - -c
        - |
          touch /data/tiflash/tmp/flash_cluster_manager.log
          tail -n0 -F /data/tiflash/tmp/flash_cluster_manager.log
        volumeMounts:
        - name: logs
          mountPath: /logs
      volumes:
      - name: logs
        emptyDir: {}
      - name: config
        configMap:
          name: tiflash
          items:
          - key: config.xml
            path: config.xml
          - key: users.xml
            path: users.xml
  volumeClaimTemplates:
  - metadata:
      name: data
    spec:
      accessModes: [ "ReadWriteOnce" ]
      storageClassName: "local-storage"
      resources:
        requests:
          storage: 10Gi
---
apiVersion: v1
kind: Service
metadata:
  name: rngine
  labels:
    app: rngine
spec:
  ports:
  - port: 20160
  clusterIP: None
  selector:
    app: rngine
---
apiVersion: v1
kind: Service
metadata:
  name: tiflash
  labels:
    app: tiflash
spec:
  ports:
  - port: 3930 # TODO: what the 8123, 9000 and 9009 port for, which type of service should we use for these service port?
  clusterIP: None
  selector:
    app: tiflash
---
apiVersion: v1
kind: ConfigMap
metadata:
  name: tiflash
data:
  rngine.toml: |
    [readpool.storage]
    [readpool.coprocessor]
    [server]
    labels = { "engine" = "tiflash", "tiflash_http_port" = "8123" }
    # TODO: does tiflash support multiple instances, how should we configure the engine-addr if there are multiple tiflash instance
    # Replace the MY_NAMESPACE with the real namespace
    engine-addr = "tiflash-0.tiflash.MY_NAMESPACE.svc:3930" # tics address
    [storage]
    [pd]
    # This section will be overwritten by command line parameters
    [metric]
    #address = "172.16.30.31:9531"
    #interval = "15s"
    #job = "tikv"
    [raftstore]
    raftdb-path = ""
    sync-log = true
    max-leader-missing-duration = "22s"
    abnormal-leader-missing-duration = "21s"
    peer-stale-state-check-interval = "20s"
    [coprocessor]
    [rocksdb]
    wal-dir = ""
    [rocksdb.defaultcf]
    block-cache-size = "10GB"
    [rocksdb.lockcf]
    block-cache-size = "4GB"
    [rocksdb.writecf]
    block-cache-size = "4GB"
    [raftdb]
    [raftdb.defaultcf]
    block-cache-size = "1GB"
    [security]
    ca-path = ""
    cert-path = ""
    key-path = ""
    [import]


  # To better support K8s, the related address needs to be configured from command line arguments.
  # eg. raft.pd_addr, flash.service_addr, flash.tidb_status_addr
  tiflash-config.xml: |
    <?xml version="1.0"?>
    <yandex>
        <application>
            <runAsDaemon>true</runAsDaemon>
        </application>

        <logger>
            <level>trace</level>
            <log>/logs/tiflash-server.log</log>
            <errorlog>/logs/tiflash-error.log</errorlog>
            <size>1000M</size>
            <count>10</count>
            <!-- <console>1</console> --> <!-- Default behavior is autodetection (log to console if not daemon mode and is tty) -->
        </logger>

        <display_name>the flash</display_name>

        <raft>
            <kvstore_path>/data/tiflash/kvstore</kvstore_path>
            <pd_addr>TIDB_CLUSTER_NAME-pd:2379</pd_addr>
            <ignore_databases>system</ignore_databases>
            <learner_key>engine</learner_key>
            <learner_value>tiflash</learner_value>
        </raft>

        <flash>
          <service_addr>tiflash-0.tiflash.MY_NAMESPACE.svc:3930</service_addr> <!--tics address-->
            <flash_cluster>
                <master_ttl>60</master_ttl>
                <refresh_interval>20</refresh_interval>
                <update_rule_interval>5</update_rule_interval>
            </flash_cluster>
            <tidb_status_addr>TIDB_CLUSTER_NAME-tidb:10080</tidb_status_addr>
        </flash>

        <http_port>8123</http_port>
        <tcp_port>9000</tcp_port>
        <interserver_http_port>9009</interserver_http_port>
        <!--listen_host>::1</listen_host-->
        <listen_host>0.0.0.0</listen_host>

        <!-- Don't exit if ipv6 or ipv4 unavailable, but listen_host with this protocol specified -->
        <!-- <listen_try>0</listen_try> -->

        <!-- Allow listen on same address:port -->
        <!-- <listen_reuse_port>0</listen_reuse_port> -->

        <!-- <listen_backlog>64</listen_backlog> -->

        <max_connections>4096</max_connections>
        <keep_alive_timeout>3</keep_alive_timeout>

        <!-- Maximum number of concurrent queries. -->
        <max_concurrent_queries>100</max_concurrent_queries>

        <!-- Set limit on number of open files (default: maximum). This setting makes sense on Mac OS X because getrlimit() fails to retrieve
             correct maximum value. -->
        <!-- <max_open_files>262144</max_open_files> -->

        <!-- Size of cache of uncompressed blocks of data, used in tables of MergeTree family.
             In bytes. Cache is single for server. Memory is allocated only on demand.
             Cache is used when 'use_uncompressed_cache' user setting turned on (off by default).
             Uncompressed cache is advantageous only for very short queries and in rare cases.
          -->
        <uncompressed_cache_size>8589934592</uncompressed_cache_size>

        <!-- Approximate size of mark cache, used in tables of MergeTree family.
             In bytes. Cache is single for server. Memory is allocated only on demand.
             You should not lower this value.
          -->
        <mark_cache_size>5368709120</mark_cache_size>

        <!-- Path of persisted mapping cache in fast(er) disk device.
             Empty means disabled.
             Seperated with ';' if there are more than one path.
          -->
        <persisted_mapping_cache_path></persisted_mapping_cache_path>
        <!-- Quota size in bytes of persisted mapping cache, in bytes. 0 means unlimited. -->
        <persisted_mapping_cache_size>0</persisted_mapping_cache_size>

        <!-- Path to data directory, with trailing slash. -->
        <path>/data/tiflash/db</path>

        <!-- Path to temporary data for processing hard queries. -->
        <tmp_path>/data/tiflash/tmp</tmp_path>

        <l0_optimize>false</l0_optimize>

        <!-- Path to configuration file with users, access rights, profiles of settings, quotas. -->
        <users_config>/etc/tiflash/users.xml</users_config>

        <!-- Default profile of settings. -->
        <default_profile>default</default_profile>

        <!-- System profile of settings. This settings are used by internal processes (Buffer storage, Distibuted DDL worker and so on). -->
        <!-- <system_profile>default</system_profile> -->

        <!-- Default database. -->
        <default_database>default</default_database>

        <!-- Server time zone could be set here.

             Time zone is used when converting between String and DateTime types,
              when printing DateTime in text formats and parsing DateTime from text,
              it is used in date and time related functions, if specific time zone was not passed as an argument.

             Time zone is specified as identifier from IANA time zone database, like UTC or Africa/Abidjan.
             If not specified, system time zone at server startup is used.

             Please note, that server could display time zone alias instead of specified name.
             Example: W-SU is an alias for Europe/Moscow and Zulu is an alias for UTC.
        -->
        <!-- <timezone></timezone> -->

        <!-- You can specify umask here (see "man umask"). Server will apply it on startup.
             Number is always parsed as octal. Default umask is 027 (other users cannot read logs, data files, etc; group can only read).
        -->
        <!-- <umask>022</umask> -->

        <!-- Reloading interval for embedded dictionaries, in seconds. Default: 3600. -->
        <builtin_dictionaries_reload_interval>3600</builtin_dictionaries_reload_interval>

        <!-- Maximum session timeout, in seconds. Default: 3600. -->
        <max_session_timeout>3600</max_session_timeout>

        <!-- Default session timeout, in seconds. Default: 60. -->
        <default_session_timeout>600</default_session_timeout>

        <!-- Query log. Used only for queries with setting log_queries = 1. -->
        <query_log>
            <!-- What table to insert data. If table is not exist, it will be created.
                 When query log structure is changed after system update,
                  then old table will be renamed and new table will be created automatically.
            -->
            <database>system</database>
            <table>query_log</table>
            <!--
                PARTITION BY expr https://clickhouse.yandex/docs/en/table_engines/custom_partitioning_key/
                Example:
                    event_date
                    toMonday(event_date)
                    toYYYYMM(event_date)
                    toStartOfHour(event_time)
            -->
            <partition_by>toYYYYMM(event_date)</partition_by>
            <!-- Interval of flushing data. -->
            <flush_interval_milliseconds>7500</flush_interval_milliseconds>
        </query_log>

        <!-- Uncomment if use part_log
        <part_log>
            <database>system</database>
            <table>part_log</table>

            <flush_interval_milliseconds>7500</flush_interval_milliseconds>
        </part_log>
        -->

        <dictionaries_config>*_dictionary.xml</dictionaries_config>

        <!-- Uncomment if you want data to be compressed 30-100% better.
             Don't do that if you just started using ClickHouse.
          -->
        <compression>
        <!--
            <!- - Set of variants. Checked in order. Last matching case wins. If nothing matches, lz4 will be used. - ->
            <case>

                <!- - Conditions. All must be satisfied. Some conditions may be omitted. - ->
                <min_part_size>10000000000</min_part_size>        <!- - Min part size in bytes. - ->
                <min_part_size_ratio>0.01</min_part_size_ratio>   <!- - Min size of part relative to whole table size. - ->

                <!- - What compression method to use. - ->
                <method>zstd</method>
            </case>
        -->
        </compression>

        <!-- Settings to fine tune MergeTree tables. See documentation in source code, in MergeTreeSettings.h -->
        <!--
        <merge_tree>
            <max_suspicious_broken_parts>5</max_suspicious_broken_parts>
        </merge_tree>
        -->

        <!-- Protection from accidental DROP.
             If size of a MergeTree table is greater than max_table_size_to_drop (in bytes) than table could not be dropped with any DROP query.
             If you want do delete one table and don't want to restart clickhouse-server, you could create special file <clickhouse-path>/flags/force_drop_table and make DROP once.
             By default max_table_size_to_drop is 50GB, max_table_size_to_drop=0 allows to DROP any tables.
             Uncomment to disable protection.
        -->
        <max_table_size_to_drop>0</max_table_size_to_drop>
    </yandex>

  tiflash-user.xml: |
    <?xml version="1.0"?>
    <yandex>
        <!-- Profiles of settings. -->
        <profiles>
            <!-- Default settings. -->
            <default>
                <!-- Maximum memory usage for processing single query, in bytes. -->
                <max_memory_usage>4000000000</max_memory_usage>

                <!-- Use cache of uncompressed blocks of data. Meaningful only for processing many of very short queries. -->
                <use_uncompressed_cache>0</use_uncompressed_cache>

                <!-- How to choose between replicas during distributed query processing.
                     random - choose random replica from set of replicas with minimum number of errors
                     nearest_hostname - from set of replicas with minimum number of errors, choose replica
                      with minumum number of different symbols between replica's hostname and local hostname
                      (Hamming distance).
                     in_order - first live replica is choosen in specified order.
                -->
                <load_balancing>random</load_balancing>
            </default>

            <!-- Profile that allows only read queries. -->
            <readonly>
                <readonly>1</readonly>
            </readonly>
        </profiles>

        <!-- Users and ACL. -->
        <users>
            <!-- If user name was not specified, 'default' user is used. -->
            <default>
                <!-- Password could be specified in plaintext or in SHA256 (in hex format).

                     If you want to specify password in plaintext (not recommended), place it in 'password' element.
                     Example: <password>qwerty</password>.
                     Password could be empty.

                     If you want to specify SHA256, place it in 'password_sha256_hex' element.
                     Example: <password_sha256_hex>65e84be33532fb784c48129675f9eff3a682b27168c0ea744b2cf58ee02337c5</password_sha256_hex>

                     How to generate decent password:
                     Execute: PASSWORD=$(base64 < /dev/urandom | head -c8); echo "$PASSWORD"; echo -n "$PASSWORD" | sha256sum | tr -d '-'
                     In first line will be password and in second - corresponding SHA256.
                -->
                <password></password>

                <!-- List of networks with open access.

                     To open access from everywhere, specify:
                        <ip>::/0</ip>

                     To open access only from localhost, specify:
                        <ip>::1</ip>
                        <ip>127.0.0.1</ip>

                     Each element of list has one of the following forms:
                     <ip> IP-address or network mask. Examples: 213.180.204.3 or 10.0.0.1/8 or 2a02:6b8::3 or 2a02:6b8::3/64.
                     <host> Hostname. Example: server01.yandex.ru.
                         To check access, DNS query is performed, and all received addresses compared to peer address.
                     <host_regexp> Regular expression for host names. Example, ^server\d\d-\d\d-\d\.yandex\.ru$
                         To check access, DNS PTR query is performed for peer address and then regexp is applied.
                         Then, for result of PTR query, another DNS query is performed and all received addresses compared to peer address.
                         Strongly recommended that regexp is ends with $
                     All results of DNS requests are cached till server restart.
                -->
                <networks>
                    <ip>::/0</ip>
                </networks>

                <!-- Settings profile for user. -->
                <profile>default</profile>

                <!-- Quota for user. -->
                <quota>default</quota>
            </default>

            <!-- Example of user with readonly access. -->
            <readonly>
                <password></password>
                <networks>
                    <ip>::1</ip>
                    <ip>127.0.0.1</ip>
                </networks>
                <profile>readonly</profile>
                <quota>default</quota>
            </readonly>
        </users>

        <!-- Quotas. -->
        <quotas>
            <!-- Name of quota. -->
            <default>
                <!-- Limits for time interval. You could specify many intervals with different limits. -->
                <interval>
                    <!-- Length of interval. -->
                    <duration>3600</duration>

                    <!-- No limits. Just calculate resource usage for time interval. -->
                    <queries>0</queries>
                    <errors>0</errors>
                    <result_rows>0</result_rows>
                    <read_rows>0</read_rows>
                    <execution_time>0</execution_time>
                </interval>
            </default>
        </quotas>
    </yandex>