Circuit Breakers

Due to the nature of Hydrolix and the potential for queries to cover vast amounts of data, the following Circuit breakers are provided to help ensure resources are consumed effectively.

The following options specify query limits, if the query goes above a specified limit it is cancelled automatically.

`hdx_query_max_rows`

Query Option	Min Value	Max Value	Default Value
`hdx_query_max_rows`	0	n/a	unlimited

Specifies the maximum number of rows that can be evaluated to answer a query

`hdx_query_max_attempts`

Query Option	Min Value	Max Value	Default Value
`hdx_query_max_attempts`	0	n/a	unlimited

Specifies the maximum number of failures that can occur

`hdx_query_max_result_bytes`

Query Option	Min Value	Max Value	Default Value
`hdx_query_max_result_bytes`	0	n/a	unlimited

The maximum amount of bytes that can be stored on the Query Head before returning data.

`hdx_query_max_result_rows`

Query Option	Min Value	Max Value	Default Value
`hdx_query_max_result_rows`	0	n/a	unlimited

The maximum number of rows that can be stored as a result of a query on the Query Head before returning the response.

`hdx_query_max_timerange_sec`

Query Option	Min Value	Max Value	Default Value
`hdx_query_max_timerange_sec`	0	n/a	unlimited

Specify the maximum timerange allowed in the query, for example 86400 will limit the timerange to 1 day. If you don't have any timerange filtering in your query this options will not work as we calculate the difference between timestamp in the WHERE clause.

`hdx_query_timerange_required`

Query Option	Possible values	Default Value
`hdx_query_timerange_required`	true / false	false

Boolean to allow a query to run if there's no time range specified in the WHERE clause. Due to the size of data that can be stored in Hydrolix this setting protects against a query forcing a scan of the whole data set.

`hdx_query_max_partitions`

Query Option	Min Value	Max Value	Default Value
`hdx_query_max_partitions`	1	n/a	unlimited

Number of partition a query is allowed to access. If that number is exceeded, the query will not be allowed to run as it'll read more partition than allowed.

`hdx_query_max_execution_time`

Query Option	Min Value	Max Value	Default Value
`max_execution_time`	0	(Unbound)	0

Number of seconds before a query is canceled. A value of 0 means that there is no limit. Any other value above 0 will cancel the query when the specified number of seconds is reached.

`hdx_query_max_columns_to_read`

Query Option	Min Value	Max Value	Default Value
`max_columns_to_read`	0	(Unbound)	0

Number of columns allowed in SELECT statement before a query is canceled. A value of 0 means that there is no limit. Any other value above 0 will cancel the query when the specified number of columns is reached.

`hdx_query_max_memory_usage`

Query Option	Min Value	Max Value	Default Value
`hdx_query_max_memory_usage`	0	(Unbound)	0

Max memory in bytes for a single query, if the query is exceeding the max_memory_usage it'll be cancel.
This settings is per query per query-peer and head.

Rate Limiting

These options specify limits on the resources available to query processes.

`hdx_query_max_peers`

Query Option	Min Value	Max Value	Default Value
`hdx_query_max_peers`	1	(Unbound)	null (all peers)

By default, Hydrolix distributes query processing across all available query peers in order to maximize massively parallel processing. Setting this flag instructs the query head to instead use only a subset of available peers. If a number greater than the number of available peers is given, all available peers are used.

`hdx_query_pool_name`

Query Option	Min Value	Max Value	Default Value
`hdx_query_pool_name`	n/a	n/a	"" (empty string)

A string with a pool name can be used to instruct Hydrolix where a query should run. Given a set of pools, using the name of a given pool will run the query only in peers belonging to the pool chosen. If the parameter is not set, the query will run in all available peers from all pools by default.

`hdx_query_max_streams`

Query Option	Min Value	Max Value	Default Value
`hdx_query_max_streams`	1	Twice the count of available CPU cores	null (1 per core)

By default, each query peer will run one process per CPU core. To limit the number of processes a query might run, set a value here.

`hdx_query_max_concurrent_partitions`

Query Option	Min Value	Max Value	Default Value
`hdx_query_max_concurrent_partitions`	1	n/a	3

Hydrolix query processing generally requires each query process to extract data from many HDX partitions. This flag sets a limit on the number of partitions which a query peer reads from at the same time.

Note that this setting is applied per-process. For example, if four-core query peer runs four processes, and each of these opens up to 25 partitions, then each query peer may have as many as 100 partitions open at once.

Decreasing this setting from its default setting of 3 may slow down query performance. Increasing this setting beyond 25 risks excessive memory pressure on the peer. Tread carefully.

Other Flags

`hdx_query_distributed_aggregation_memory_efficient`

Query Option	Min Value	Max Value	Default Value
`hdx_query_distributed_aggregation_memory_efficient`	0	n/a	unlimited

Relates to https://github.com/ClickHouse/ClickHouse/pull/20599

`hdx_query_max_bytes_before_external_group_by`

Query Option	Min Value	Max Value	Default Value
`hdx_query_max_bytes_before_external_group_by`	0	n/a	unlimited

The maximum amount of bytes that can be used in memory before data is spilt to disk when applying group bys. These can be used to help protect out of memory errors, however disk will be utilized in replacement. Use with care.

`hdx_query_max_bytes_before_external_sort`

Query Option	Min Value	Max Value	Default Value
`hdx_query_max_bytes_before_external_sort`	0	n/a	unlimited

The maximum amount of bytes that can be used in memory before data is spilt to disk when applying sort bys. These can be used to help protect out of memory errors, however disk will be utilized in replacement. Use with care.

`hdx_query_unlimited_cnf`

Query Option	Min Value	Max Value	Default Value
`hdx_query_unlimited_cnf`	0	1	0

When set to 1, this disables limits on the number of clauses when converting the query to conjunctive normal form (CNF). See the ClickHouse documentation for more information. Note that disabling this cap on CNFs will likely cause the query to be very slow and potentially use much more memory.