0
0
mirror of https://github.com/PostHog/posthog.git synced 2024-12-01 04:12:23 +01:00
posthog/hogvm/README.md

5.1 KiB

HogVM

A HogVM is a 🦔 that runs HogQL bytecode. It's purpose is to locally evaluate HogQL expressions against any object.

HogQL bytecode

HogQL Bytecode is a compact representation of a subset of the HogQL AST nodes. It follows a certain structure:

1 + 2                  # [_H, op.INTEGER, 2, op.INTEGER, 1, op.PLUS]
1 and 2                # [_H, op.INTEGER, 2, op.INTEGER, 1, op.AND, 2]
1 or 2                 # [_H, op.INTEGER, 2, op.INTEGER, 1, op.OR, 2]
not true               # [_H, op.TRUE, op.NOT]
properties.bla         # [_H, op.STRING, "bla", op.STRING, "properties", op.FIELD, 2]
call('arg', 'another') # [_H, op.STRING, "another", op.STRING, "arg", op.CALL, "call", 2]
1 = 2                  # [_H, op.INTEGER, 2, op.INTEGER, 1, op.EQ]
'bla' !~ 'a'           # [_H, op.STRING, 'a', op.STRING, 'bla', op.NOT_REGEX]

Compliant implementation

The python/execute.py function in this folder acts as the reference implementation in case of disputes.

Operations

To be considered a PostHog HogQL Bytecode Certified Parser, you must implement the following operations:

FIELD = 1          # [arg3, arg2, arg1, FIELD, 3]       # arg1.arg2.arg3
CALL = 2           # [arg2, arg1, CALL, 'concat', 2]    # concat(arg1, arg2)
AND = 3            # [val3, val2, val1, AND, 3]         # val1 and val2 and val3
OR = 4             # [val3, val2, val1, OR, 3]          # val1 or val2 or val3
NOT = 5            # [val, NOT]                         # not val
PLUS = 6           # [val2, val1, PLUS]                 # val1 + val2
MINUS = 7          # [val2, val1, MINUS]                # val1 - val2
MULTIPLY = 8       # [val2, val1, MULTIPLY]             # val1 * val2
DIVIDE = 9         # [val2, val1, DIVIDE]               # val1 / val2
MOD = 10           # [val2, val1, MOD]                  # val1 % val2
EQ = 11            # [val2, val1, EQ]                   # val1 == val2
NOT_EQ = 12        # [val2, val1, NOT_EQ]               # val1 != val2
GT = 13            # [val2, val1, GT]                   # val1 > val2
GT_EQ = 14         # [val2, val1, GT_EQ]                # val1 >= val2
LT = 15            # [val2, val1, LT]                   # val1 < val2
LT_EQ = 16         # [val2, val1, LT_EQ]                # val1 <= val2
LIKE = 17          # [val2, val1, LIKE]                 # val1 like val2
ILIKE = 18         # [val2, val1, ILIKE]                # val1 ilike val2
NOT_LIKE = 19      # [val2, val1, NOT_LIKE]             # val1 not like val2
NOT_ILIKE = 20     # [val2, val1, NOT_ILIKE]            # val1 not ilike val2
IN = 21            # [val2, val1, IN]                   # val1 in val2
NOT_IN = 22        # [val2, val1, NOT_IN]               # val1 not in val2
REGEX = 23         # [val2, val1, REGEX]                # val1 =~ val2
NOT_REGEX = 24     # [val2, val1, NOT_REGEX]            # val1 !~ val2
IREGEX = 25        # [val2, val1, IREGEX]               # val1 =~* val2
NOT_IREGEX = 26    # [val2, val1, NOT_IREGEX]           # val1 !~* val2
TRUE = 29          # [TRUE]                             # true
FALSE = 30         # [FALSE]                            # false
NULL = 31          # [NULL]                             # null
STRING = 32        # [STRING, 'text']                   # 'text'
INTEGER = 33       # [INTEGER, 123]                     # 123
FLOAT = 34         # [FLOAT, 123.12]                    # 123.01

Async Operations

Some operations can't be computed directly, and are thus asked back to the caller. These include:

IN_COHORT = 27     # [val2, val1, IREGEX]               # val1 in cohort val2
NOT_IN_COHORT = 28 # [val2, val1, NOT_IREGEX]           # val1 not in cohort val2

The arguments for these instructions will be passed on to the provided async_operation(*args) in reverse:

def async_operation(*args):
    if args[0] == op.IN_COHORT:
        return db.queryInCohort(args[1], args[2])
    return False

execute_bytecode(to_bytecode("'user_id' in cohort 2"), {}, async_operation)

Functions

A PostHog HogQL Bytecode Certified Parser must also implement the following function calls:

concat(...)             # concat('test: ', 1, null, '!') == 'test: 1!'
match(string, pattern)  # match('fish', '$fi.*') == true
toString(val)           # toString(true) == 'true'
toInt(val)              # toInt('123') == 123
toFloat(val)            # toFloat('123.2') == 123.2
toUUID(val)             # toUUID('string') == 'string'

Null handling

In HogQL equality comparisons, null is treated as any other variable. Its presence will not make functions automatically return null, as is the ClickHouse default.

1 == null # false
1 != null # true

Nulls are just ignored in concat

Known broken features

  • Regular Expression support is implemented, but NOT GUARANTEED to the same way across platforms. Different implementations (ClickHouse, Python, Node) use different Regexp engines. ClickHouse uses re2, the others use pcre. Use the case-insensitive regex operators instead of passing in modifier flags through the expression.
  • DateTime comparisons are not supported.
  • Cohort Matching operations are not implemented.
  • Only a small subset of functions is enabled. This list is bound to expand.