Troubleshoot issues with Zookeeper znodes.
Pinot stores cluster metadata, including schema definitions, table configuration, and segment assignment, in ZooKeeper. Internally, Zookeeper uses a hierarchical namespace, similar to a file system. The nodes in this hierarchy are called znodes and they can store data and have child nodes.
The default maximum size of znodes is 1MB, which is sufficient for most deployments. However, if you have 100s of thousands of segments, it's possible that this size limit is exceeded. If this happens, you will see an error message similar to the following:
To address the size limit is exceeded error, do the following:
Reduce the number of segments
Adjust ZooKeeper znode size
Reduce the number of segments to reduce the metadata stored in the IDEALSTATE
and EXTERNALVIEW
znodes, which are the two znodes most likely to exceed 1MB.
To do this for new segments, configure the segment threshold to a higher value. For existing segments, run the Minion merge rollup task.
Adjust the maximum size of znodes in ZooKeeper. To do this, configure the jute.maxbuffer
Java system property, which defines the maximum znode size in bytes. To read more about this property, see the ZooKeeper documentation.
We recommend setting this value to 4MB. Set this parameter to the same value on the ZooKeeper node and clients that are interacting with ZooKeeper (all Pinot components).
Troubleshoot common issues in Apache Pinot.
Troubleshoot issues with the multi-stage query engine (v2).
Learn how to troubleshoot errors when using the multi-stage query engine (v2), and see multi-stage query engine limitations.
Find instructions on how to enable the multi-stage query engine, or see a high-level overview of how the multi-stage query engine works.
We are continuously improving the multi-stage query engine. A few limitations to call out:
Support for multi-value columns is limited to projections, and predicates must use the arrayToMv
function. For example, to successfully run the following query:
You must include arrayToMv
in the query as follows:
Schema and other prefixes are not supported in queries. For example, the following queries are not supported:
Queries without prefixes are supported:
Modifying query behavior based on the cluster configuration is not supported. distinctcounthll
, distinctcounthllmv
, distinctcountrawhll
, and `distinctcountrawhllmv` use
a different default value of log2mParam
in the multi-stage engine. In multi-stage, this value can no longer be configured. Therefore, the following query may produce different results in single-stage and multi-stage engine:
To ensure multi-stage returns the same result, specify the log2mParam
value in your query:
If a column is repeated more than once in SELECT statement, that column requires disambiguate aliasing. For example, in the following query, the reference to colA
is ambiguous whether it's to the first or second projected colA
:
The solution is to rewrite the query either use aliasing:
Or use index-based referencing:
Pinot single-stage query engine automatically removes the underscore _ character from function names. So co_u_n_t()
is equivalent to count().
In multi-stage, function naming restrictions were tightened, so the underscore(_)
character is only allowed to separate word boundaries in a function name. Also camel case is supported in function names. For example, the following function names are allowed:
Pinot single-stage query engine automatically do implicit type casts in many of the situations, for example when running the following:
it will automatically convert both values to long datatypes before comparison. This behavior however could cause issues and thus it is not so widely applied in the multi-stage engine where a stricter datatype conformance is enforced. the example above should be explicitly written as:
Default names for projections with function calls are different between single and multi-stage.
For example, in multi-stage, the following query:
Returns the following result:
In single-stage, the following function:
Returns the following result:
In multi-stage, table and column names and are case sensitive. In single-stage they were not. For example, the following two queries are not equivalent in multi-stage engine:
select * from myTable
select * from mytable
Note: Function names are not case sensitive in neither single nor multi-stage.
An arbitrary number of arguments is no longer supported in multi-stage. For example, in single-stage, the following query worked:
In multi-stage, this query must be rewritten as follows:
Note: Remember that select 1 + 2 + 3 + 4 + 5 from table
is still valid in multi-stage
Null handling is not supported when tables use table based null storing. You have to use column base null storing instead. See null handling support
In multi-stage:
The histogram
function is not supported.
The timeConvert
function is not supported, see dateTimeConvert
for more details.
The dateTimeConvertWindowHop
function is not supported.
Array & Map-related functions are not supported.
aggregate function that requires literal input (such as percentile
, firstWithTime
) might result in a non-compilable query plan.
The multi-stage engine uses different type names than the single-stage engine. Although the classical names must still be used in schemas and some SQL expressions, the new names must be used in CAST expressions.
The following table shows the differences in type names:
VARBINARY literals in multi-stage engine must be prefixed with X
or x
. For example, the following query:
In single-stage engine the same query would be:
Troubleshoot semantic/runtime errors and timeout errors.
Try downloading the latest docker image or building from the latest master commit.
We continuously push bug fixes for the multi-stage engine so bugs you encountered might have already been fixed in the latest master build.
Try rewriting your query.
Some functions previously supported in the single-stage query engine (v1) may have a new way to express in the multi-stage engine (v2). Check and see if you are using any non-standard SQL functions or semantics.
Try reducing the size of the table(s) used.
Add higher selectivity filters to the tables.
Try executing part of the subquery or a simplified version of the query first.
This helps to determine the selectivity and scale of the query being executed.
Try adding more servers.
The new multi-stage engine runs distributed across the entire cluster, so adding more servers to partitioned queries such as GROUP BY aggregates, and equality JOINs help speed up the query runtime.
Single-stage engine | Multi-stage engine |
---|---|
NULL
NULL
BOOLEAN
BOOLEAN
INT
INT
LONG
BIGINT
BIG_DECIMAL
DECIMAL
FLOAT
FLOAT/REAL
DOUBLE
DOUBLE
INTERVAL
INTERVAL
TIMESTAMP
TIMESTAMP
STRING
VARCHAR
BYTES
VARBINARY
-
ARRAY
JSON
-