--mode="ingestor": Operation mode. You can choose between ingestor and store modes.
In store mode, seq-db acts as a stateful replica responsible for storing its part of the data. Implements an internal API that is used to communicate between the store and ingestor.
In ingestor mode, seq-db acts as a shard and replica coordinator. Implements a public client API. Does not have its own state.
--addr=":9002": Depending on the mode:
For ingestor mode, this is the address of the public HTTP API (bulk method).
For store mode, this is the address of the internal gRPC API. By default, port 9002 is used.
--debug-addr=":9200": Address for debugging requests (HTTP). Go metrics and profiling are sent to this address. By default, port 9200 is used.
--mapping=MAPPING: Path to the file with indexing parameters or value auto. See the corresponding section.
--case-sensitive: Token case sensitivity. By default, if not specified, the search is case-insensitive.
--max-token-size=72: Maximum token size.
--partial-indexing: By default, if the indexed value exceeds the maximum size, it is ignored and does not get into the index. If this parameter is set, the value will be truncated to the maximum length and indexed.
--write-stores=WRITE-STORES: List of hosts for writing data. Specified as a string with values separated by commas. For example, --write-stores=host1,host2,host3,host4.
--read-stores=READ-STORES: List of hosts for reading data. If not set, --write-stores is used. Can be used, for example, in case of data migration to other hosts, when we write to some stores and read from other, old stores.
--hot-stores=HOT-STORES: List of hot storage hosts. If specified, the proxy works with 2 store clusters: cold (--write-stores) and hot (--hot-stores). And sends each write request to each of these clusters accordingly. But when reading, it first tries to get data from the hot cluster and in some cases from the cold one.
--hot-read-stores=HOT-READ-STORES: List of hot storage hosts for reading. Can be used when migrating data to other hosts.
--store-mode="": Storage operating mode. If specified, the allowed values are hot or cold. If the hot mode is selected, then when executing a search query, if the requested data range is older than the oldest store documents, the service will return a special error query wants old data. Ingestor will make a request to the cold store cluster in this case.
--replicas=1: Replication factor for storages. If N is specified, the first N hosts in the list are replicas of one shard, the next N are replicas of another shard, etc.
--hot-replicas=HOT-REPLICAS: Replication factor for hot storages. If not specified, the global factor is used.
--shuffle-replicas: Shuffle replicas before performing a search. If not specified, then the first replica is always read, and only in case of failure, the second one comes.
--bulk-shard-timeout=10s: Timeout for processing a bulk operation by one shard.
--bulk-err-percentage=50: Error percentage for triggering the overload protection mechanism. See circuitbreaker/README.md for more details.
--bulk-bucket-width=1s: Window width for counting errors. See circuitbreaker/README.md for more details.
--bulk-err-count=10: Number of errors required to trigger bulk protection mechanism. See circuitbreaker/README.md for more details.
--bulk-sleep-window=5s: Time to wait after bulk protection mechanism is triggered. See circuitbreaker/README.md for more details.
--bulk-request-volume-threshold=5: Request volume threshold for bulk protection mechanism to be triggered. See circuitbreaker/README.md for more details.
--max-inflight-bulks=32: Maximum number of concurrent bulk requests that can be processed.
--query-rate-limit=2.0: Maximum request rate per second. Search and fetch requests are counted. If the limit is exceeded, the request will return an error.
--data-dir=DATA-DIR: Directory where data is stored.
--frac-size=128MB: Size of one fraction (minimum fragment of data on disk). The larger the fraction size, the more RAM is required for fraction buffering.
--total-size=1GB: Maximum size of all data. If the data becomes larger than this limit, the oldest data is deleted until the dataset size fits within the specified value.
--reader-workers=128: The size of the reader pool from the disk. For SSDs, it is recommended to set it equal to the number of cores.
--search-workers-count=128: The number of worker threads that will process the search by fraction. It is recommended to set it equal to the number of cores.
--search-requests-limit=30: The maximum number of simultaneous search requests. If exceeded, the request will return an error.
--search-fraction-limit=6000: The maximum number of fractions used in the search. If the query requires reading more fractions, the query will return an error.
--max-search-docs=100000: The maximum number of documents returned by the search query.
--cache-size=8GB: The maximum cache size. Used when processing search and fetch requests.
--agg-max-group-tokens=2000: Maximum number of unique tokens for a grouping field affected by an aggregation query. Setting this value to 0 disables the limitation.
--agg-max-field-tokens=1000000: Maximum number of unique tokens for an aggregation function field affected by an aggregation query. Setting this value to 0 disables the limitation.
--agg-max-fraction-tids=100000: Maximum number of unique tokens per fraction for an grouping field or for an aggregation function calculation field. Setting this value to 0 disables the limitation.
--log-search-threshold-ms=3000:Search query logging threshold in milliseconds. All queries that take longer than the specified time are written to the log.
--log-fetch-threshold-ms=3000:Fetch query logging threshold in milliseconds. All queries that take longer than the specified time are written to the log.
--log-bulk-threshold-ms=LOG-BULK-THRESHOLD-MS:Bulk query logging threshold in milliseconds. A record of such a query is written to the log.