def es_setop(es, query):
schema = query.frum.schema
query_path = schema.query_path[0]
new_select, split_select = get_selects(query)
split_wheres = split_expression_by_path(query.where, schema, lang=ES52)
es_query = es_query_proto(query_path, split_select, split_wheres, schema)
es_query.size = coalesce(query.limit, DEFAULT_LIMIT)
es_query.sort = jx_sort_to_es_sort(query.sort, schema)
with Timer("call to ES", silent=DEBUG) as call_timer:
result = es.search(es_query)
# Log.note("{{result}}", result=result)
T = result.hits.hits
try:
formatter, _, mime_type = set_formatters[query.format]
with Timer("formatter", silent=True):
output = formatter(T, new_select, query)
output.meta.timing.es = call_timer.duration
output.meta.content_type = mime_type
output.meta.es_query = es_query
return output
except Exception as e:
Log.error("problem formatting", e)
def extract_aggs(select, query_path, schema):
"""
RETURN ES AGGREGATIONS
"""
new_select = Data(
) # MAP FROM canonical_name (USED FOR NAMES IN QUERY) TO SELECT MAPPING
formula = []
for s in select:
if is_op(s.value, Variable_):
s.query_path = query_path
if s.aggregate == "count":
new_select["count_" + literal_field(s.value.var)] += [s]
else:
new_select[literal_field(s.value.var)] += [s]
elif s.aggregate:
op, split_select = split_expression_by_path(s.value,
schema,
lang=Painless)
for si_key, si_value in split_select.items():
if si_value:
if s.query_path:
Log.error(
"can not handle more than one depth per select")
s.query_path = si_key
formula.append(s)
acc = Aggs()
agg_field(acc, new_select, query_path, schema)
agg_formula(acc, formula, query_path, schema)
return acc
def build_es_query(select, query_path, schema, query):
acc = extract_aggs(select, query_path, schema)
acc = NestedAggs(query_path).add(acc)
split_decoders = get_decoders_by_path(query)
split_wheres = split_expression_by_path(query.where,
schema=schema,
lang=ES52)
start = 0
decoders = [None] * (len(query.edges) + len(query.groupby))
paths = list(
reversed(sorted(set(split_wheres.keys())
| set(split_decoders.keys()))))
for path in paths:
decoder = split_decoders.get(path, Null)
where = split_wheres.get(path, Null)
for d in decoder:
decoders[d.edge.dim] = d
acc = d.append_query(path, acc)
start += d.num_columns
if where:
acc = FilterAggs("_filter", AndOp(where), None).add(acc)
acc = NestedAggs(path).add(acc)
acc = NestedAggs('.').add(acc)
acc = simplify(acc)
es_query = wrap(acc.to_es(schema))
es_query.size = 0
return acc, decoders, es_query
def append_query(self, query_path, es_query):
domain = self.domain
domain_key = domain.key
value = Painless[self.edge.value]
cnv = pull_functions[value.type]
include = tuple(cnv(p[domain_key]) for p in domain.partitions)
schema = self.schema
exists = Painless[AndOp([InOp([value,
Literal(include)])])].partial_eval()
limit = coalesce(self.limit, len(domain.partitions))
if is_op(value, Variable):
es_field = first(schema.leaves(
value.var)).es_column # ALREADY CHECKED THERE IS ONLY ONE
match = TermsAggs(
"_match", {
"field": es_field,
"size": limit,
"order": {
"_term": self.sorted
} if self.sorted else None
}, self)
else:
match = TermsAggs("_match", {
"script": text(value.to_es_script(schema)),
"size": limit
}, self)
output = Aggs().add(
FilterAggs("_filter", exists, None).add(match.add(es_query)))
if self.edge.allowNulls:
# IF ALL NESTED COLUMNS ARE NULL, DOES THE FILTER PASS?
# MISSING AT THE QUERY DEPTH
op, split = split_expression_by_path(NotOp(exists), schema)
for i, p in enumerate(reversed(sorted(split.keys()))):
e = split.get(p)
if e:
not_match = NestedAggs(p).add(
FilterAggs("_missing" + text(i), e,
self).add(es_query))
output.add(not_match)
return output
def es_bulksetop(esq, frum, query):
abs_limit = MIN([query.limit, MAX_DOCUMENTS])
guid = Random.base64(32, extra="-_")
schema = query.frum.schema
query_path = schema.query_path[0]
new_select, split_select = get_selects(query)
split_wheres = split_expression_by_path(query.where, schema, lang=ES52)
es_query = es_query_proto(query_path, split_select, split_wheres, schema)
es_query.size = MIN([query.chunk_size, MAX_CHUNK_SIZE])
es_query.sort = jx_sort_to_es_sort(query.sort, schema)
if not es_query.sort:
es_query.sort = ["_doc"]
formatter = formatters[query.format](abs_limit, new_select, query)
Thread.run(
"Download " + guid,
extractor,
guid,
abs_limit,
esq,
es_query,
formatter,
parent_thread=Null,
).release()
output = wrap({
"url": URL_PREFIX / (guid + ".json"),
"status": URL_PREFIX / (guid + ".status.json"),
"meta": {
"format": query.format,
"es_query": es_query,
"limit": abs_limit
},
})
return output
def get_selects(query):
schema = query.frum.schema
split_select = {".": ESSelect(".")}
def get_select(path):
es_select = split_select.get(path)
if not es_select:
es_select = split_select[path] = ESSelect(path)
return es_select
selects = wrap([unwrap(s.copy()) for s in listwrap(query.select)])
new_select = FlatList()
put_index = 0
for select in selects:
# IF THERE IS A *, THEN INSERT THE EXTRA COLUMNS
if is_op(select.value, LeavesOp) and is_op(select.value.term, Variable):
term = select.value.term
leaves = schema.leaves(term.var)
for c in leaves:
full_name = concat_field(
select.name, relative_field(untype_path(c.name), term.var)
)
if c.jx_type == NESTED:
get_select(".").set_op = True
new_select.append(
{
"name": full_name,
"value": Variable(c.es_column),
"put": {
"name": literal_field(full_name),
"index": put_index,
"child": ".",
},
"pull": get_pull_source(c.es_column),
}
)
put_index += 1
else:
get_select(c.nested_path[0]).fields.append(c.es_column)
new_select.append(
{
"name": full_name,
"value": Variable(c.es_column),
"put": {
"name": literal_field(full_name),
"index": put_index,
"child": ".",
},
}
)
put_index += 1
elif is_op(select.value, Variable):
s_column = select.value.var
if s_column == ".":
# PULL ALL SOURCE
get_select(".").set_op = True
new_select.append(
{
"name": select.name,
"value": select.value,
"put": {"name": select.name, "index": put_index, "child": "."},
"pull": get_pull_source("."),
}
)
continue
leaves = schema.leaves(s_column) # LEAVES OF OBJECT
# nested_selects = {}
if leaves:
if any(c.jx_type == NESTED for c in leaves):
# PULL WHOLE NESTED ARRAYS
get_select(".").set_op = True
for c in leaves:
if (
len(c.nested_path) == 1
): # NESTED PROPERTIES ARE IGNORED, CAPTURED BY THESE FIRST LEVEL PROPERTIES
pre_child = join_field(
decode_property(n) for n in split_field(c.name)
)
new_select.append(
{
"name": select.name,
"value": Variable(c.es_column),
"put": {
"name": select.name,
"index": put_index,
"child": untype_path(
relative_field(pre_child, s_column)
),
},
"pull": get_pull_source(c.es_column),
}
)
else:
# PULL ONLY WHAT'S NEEDED
for c in leaves:
c_nested_path = c.nested_path[0]
if c_nested_path == ".":
if c.es_column == "_id":
new_select.append(
{
"name": select.name,
"value": Variable(c.es_column),
"put": {
"name": select.name,
"index": put_index,
"child": ".",
},
"pull": lambda row: row._id,
}
)
elif c.jx_type == NESTED:
get_select(".").set_op = True
pre_child = join_field(
decode_property(n) for n in split_field(c.name)
)
new_select.append(
{
"name": select.name,
"value": Variable(c.es_column),
"put": {
"name": select.name,
"index": put_index,
"child": untype_path(
relative_field(pre_child, s_column)
),
},
"pull": get_pull_source(c.es_column),
}
)
else:
get_select(c_nested_path).fields.append(c.es_column)
pre_child = join_field(
decode_property(n) for n in split_field(c.name)
)
new_select.append(
{
"name": select.name,
"value": Variable(c.es_column),
"put": {
"name": select.name,
"index": put_index,
"child": untype_path(
relative_field(pre_child, s_column)
),
},
}
)
else:
es_select = get_select(c_nested_path)
es_select.fields.append(c.es_column)
child = relative_field(
untype_path(
relative_field(c.name, schema.query_path[0])
),
s_column,
)
pull = accumulate_nested_doc(
c_nested_path,
Variable(
relative_field(s_column, unnest_path(c_nested_path))
),
)
new_select.append(
{
"name": select.name,
"value": select.value,
"put": {
"name": select.name,
"index": put_index,
"child": child,
},
"pull": pull,
}
)
else:
new_select.append(
{
"name": select.name,
"value": Variable("$dummy"),
"put": {"name": select.name, "index": put_index, "child": "."},
}
)
put_index += 1
else:
split_scripts = split_expression_by_path(
select.value, schema, lang=Painless
)
for p, script in split_scripts.items():
es_select = get_select(p)
es_select.scripts[select.name] = {
"script": text(
Painless[first(script)].partial_eval().to_es_script(schema)
)
}
new_select.append(
{
"name": select.name,
"pull": jx_expression_to_function(
"fields." + literal_field(select.name)
),
"put": {"name": select.name, "index": put_index, "child": "."},
}
)
put_index += 1
for n in new_select:
if n.pull:
continue
elif is_op(n.value, Variable):
if get_select(".").set_op:
n.pull = get_pull_source(n.value.var)
elif n.value == "_id":
n.pull = jx_expression_to_function("_id")
else:
n.pull = jx_expression_to_function(
concat_field("fields", literal_field(n.value.var))
)
else:
Log.error("Do not know what to do")
return new_select, split_select
def es_setop(es, query):
schema = query.frum.schema
query_path = schema.query_path[0]
split_select = {".": ESSelect('.')}
def get_select(path):
es_select = split_select.get(path)
if not es_select:
es_select = split_select[path] = ESSelect(path)
return es_select
selects = wrap([unwrap(s.copy()) for s in listwrap(query.select)])
new_select = FlatList()
put_index = 0
for select in selects:
# IF THERE IS A *, THEN INSERT THE EXTRA COLUMNS
if is_op(select.value, LeavesOp) and is_op(select.value.term, Variable):
term = select.value.term
leaves = schema.leaves(term.var)
for c in leaves:
full_name = concat_field(select.name, relative_field(untype_path(c.name), term.var))
if c.jx_type == NESTED:
get_select('.').use_source = True
new_select.append({
"name": full_name,
"value": Variable(c.es_column),
"put": {"name": literal_field(full_name), "index": put_index, "child": "."},
"pull": get_pull_source(c.es_column)
})
put_index += 1
else:
get_select(c.nested_path[0]).fields.append(c.es_column)
new_select.append({
"name": full_name,
"value": Variable(c.es_column),
"put": {"name": literal_field(full_name), "index": put_index, "child": "."}
})
put_index += 1
elif is_op(select.value, Variable):
s_column = select.value.var
if s_column == ".":
# PULL ALL SOURCE
get_select('.').use_source = True
new_select.append({
"name": select.name,
"value": select.value,
"put": {"name": select.name, "index": put_index, "child": "."},
"pull": get_pull_source(".")
})
continue
leaves = schema.leaves(s_column) # LEAVES OF OBJECT
# nested_selects = {}
if leaves:
if any(c.jx_type == NESTED for c in leaves):
# PULL WHOLE NESTED ARRAYS
get_select('.').use_source = True
for c in leaves:
if len(c.nested_path) == 1: # NESTED PROPERTIES ARE IGNORED, CAPTURED BY THESE FIRST LEVEL PROPERTIES
pre_child = join_field(decode_property(n) for n in split_field(c.name))
new_select.append({
"name": select.name,
"value": Variable(c.es_column),
"put": {"name": select.name, "index": put_index, "child": untype_path(relative_field(pre_child, s_column))},
"pull": get_pull_source(c.es_column)
})
else:
# PULL ONLY WHAT'S NEEDED
for c in leaves:
c_nested_path = c.nested_path[0]
if c_nested_path == ".":
if c.es_column == "_id":
new_select.append({
"name": select.name,
"value": Variable(c.es_column),
"put": {"name": select.name, "index": put_index, "child": "."},
"pull": lambda row: row._id
})
elif c.jx_type == NESTED:
get_select('.').use_source = True
pre_child = join_field(decode_property(n) for n in split_field(c.name))
new_select.append({
"name": select.name,
"value": Variable(c.es_column),
"put": {"name": select.name, "index": put_index, "child": untype_path(relative_field(pre_child, s_column))},
"pull": get_pull_source(c.es_column)
})
else:
get_select(c_nested_path).fields.append(c.es_column)
pre_child = join_field(decode_property(n) for n in split_field(c.name))
new_select.append({
"name": select.name,
"value": Variable(c.es_column),
"put": {"name": select.name, "index": put_index, "child": untype_path(relative_field(pre_child, s_column))}
})
else:
es_select = get_select(c_nested_path)
es_select.fields.append(c.es_column)
child = relative_field(untype_path(relative_field(c.name, schema.query_path[0])), s_column)
pull = accumulate_nested_doc(c_nested_path, Variable(relative_field(s_column, unnest_path(c_nested_path))))
new_select.append({
"name": select.name,
"value": select.value,
"put": {
"name": select.name,
"index": put_index,
"child": child
},
"pull": pull
})
else:
new_select.append({
"name": select.name,
"value": Variable("$dummy"),
"put": {"name": select.name, "index": put_index, "child": "."}
})
put_index += 1
else:
split_scripts = split_expression_by_path(select.value, schema, lang=Painless)
for p, script in split_scripts.items():
es_select = get_select(p)
es_select.scripts[select.name] = {"script": text_type(Painless[first(script)].partial_eval().to_es_script(schema))}
new_select.append({
"name": select.name,
"pull": jx_expression_to_function("fields." + literal_field(select.name)),
"put": {"name": select.name, "index": put_index, "child": "."}
})
put_index += 1
for n in new_select:
if n.pull:
continue
elif is_op(n.value, Variable):
if get_select('.').use_source:
n.pull = get_pull_source(n.value.var)
elif n.value == "_id":
n.pull = jx_expression_to_function("_id")
else:
n.pull = jx_expression_to_function(concat_field("fields", literal_field(n.value.var)))
else:
Log.error("Do not know what to do")
split_wheres = split_expression_by_path(query.where, schema, lang=ES52)
es_query = es_query_proto(query_path, split_select, split_wheres, schema)
es_query.size = coalesce(query.limit, DEFAULT_LIMIT)
es_query.sort = jx_sort_to_es_sort(query.sort, schema)
with Timer("call to ES", silent=True) as call_timer:
data = es_post(es, es_query, query.limit)
T = data.hits.hits
# Log.note("{{output}}", output=T)
try:
formatter, groupby_formatter, mime_type = format_dispatch[query.format]
with Timer("formatter", silent=True):
output = formatter(T, new_select, query)
output.meta.timing.es = call_timer.duration
output.meta.content_type = mime_type
output.meta.es_query = es_query
return output
except Exception as e:
Log.error("problem formatting", e)
def agg_formula(acc, formula, query_path, schema):
# DUPLICATED FOR SCRIPTS, MAYBE THIS CAN BE PUT INTO A LANGUAGE?
for i, s in enumerate(formula):
canonical_name = s.name
s_path = [
k for k, v in split_expression_by_path(
s.value, schema=schema, lang=Painless).items() if v
]
if len(s_path) == 0:
# FOR CONSTANTS
nest = NestedAggs(query_path)
acc.add(nest)
elif len(s_path) == 1:
nest = NestedAggs(first(s_path))
acc.add(nest)
else:
raise Log.error("do not know how to handle")
if is_op(s.value, TupleOp):
if s.aggregate == "count":
# TUPLES ALWAYS EXIST, SO COUNTING THEM IS EASY
s.pull = jx_expression_to_function("doc_count")
elif s.aggregate in ('max', 'maximum', 'min', 'minimum'):
if s.aggregate in ('max', 'maximum'):
dir = 1
op = "max"
else:
dir = -1
op = 'min'
nully = Painless[TupleOp(
[NULL] *
len(s.value.terms))].partial_eval().to_es_script(schema)
selfy = text(
Painless[s.value].partial_eval().to_es_script(schema))
script = {
"scripted_metric": {
'init_script':
'params._agg.best = ' + nully + '.toArray();',
'map_script':
'params._agg.best = ' + expand_template(
MAX_OF_TUPLE, {
"expr1": "params._agg.best",
"expr2": selfy,
"dir": dir,
"op": op
}) + ";",
'combine_script':
'return params._agg.best',
'reduce_script':
'return params._aggs.stream().' + op + '(' +
expand_template(COMPARE_TUPLE, {
"dir": dir,
"op": op
}) + ').get()',
}
}
nest.add(
NestedAggs(query_path).add(
ExprAggs(canonical_name, script, s)))
s.pull = jx_expression_to_function("value")
else:
Log.error("{{agg}} is not a supported aggregate over a tuple",
agg=s.aggregate)
elif s.aggregate == "count":
nest.add(
ExprAggs(
canonical_name, {
"value_count": {
"script":
text(Painless[s.value].partial_eval().to_es_script(
schema))
}
}, s))
s.pull = jx_expression_to_function("value")
elif s.aggregate == "median":
# ES USES DIFFERENT METHOD FOR PERCENTILES THAN FOR STATS AND COUNT
key = literal_field(canonical_name + " percentile")
nest.add(
ExprAggs(
key, {
"percentiles": {
"script": text(
Painless[s.value].to_es_script(schema)),
"percents": [50]
}
}, s))
s.pull = jx_expression_to_function(join_field(["50.0"]))
elif s.aggregate in ("and", "or"):
key = literal_field(canonical_name + " " + s.aggregate)
op = aggregates[s.aggregate]
nest.add(
ExprAggs(
key, {
op: {
"script":
text(Painless[NumberOp(
s.value)].to_es_script(schema))
}
}, s))
# get_name = concat_field(canonical_name, "value")
s.pull = jx_expression_to_function({
"case": [{
"when": {
"eq": {
"value": 1
}
},
"then": True
}, {
"when": {
"eq": {
"value": 0
}
},
"then": False
}]
})
elif s.aggregate == "percentile":
# ES USES DIFFERENT METHOD FOR PERCENTILES THAN FOR STATS AND COUNT
key = literal_field(canonical_name + " percentile")
percent = mo_math.round(s.percentile * 100, decimal=6)
nest.add(
ExprAggs(
key, {
"percentiles": {
"script": text(
Painless[s.value].to_es_script(schema)),
"percents": [percent]
}
}, s))
s.pull = jx_expression_to_function(
join_field(["values", text(percent)]))
elif s.aggregate == "cardinality":
# ES USES DIFFERENT METHOD FOR CARDINALITY
key = canonical_name + " cardinality"
nest.add(
ExprAggs(
key, {
"cardinality": {
"script": text(
Painless[s.value].to_es_script(schema))
}
}, s))
s.pull = jx_expression_to_function("value")
elif s.aggregate == "stats":
# REGULAR STATS
nest.add(
ExprAggs(
canonical_name, {
"extended_stats": {
"script": text(
Painless[s.value].to_es_script(schema))
}
}, s))
s.pull = get_pull_stats()
# GET MEDIAN TOO!
select_median = s.copy()
select_median.pull = jx_expression_to_function(
{"select": [{
"name": "median",
"value": "values.50\\.0"
}]})
nest.add(
ExprAggs(
canonical_name + "_percentile", {
"percentiles": {
"script": text(
Painless[s.value].to_es_script(schema)),
"percents": [50]
}
}, select_median))
s.pull = get_pull_stats()
elif s.aggregate == "union":
# USE TERMS AGGREGATE TO SIMULATE union
nest.add(
TermsAggs(canonical_name, {
"script_field":
text(Painless[s.value].to_es_script(schema))
}, s))
s.pull = jx_expression_to_function("key")
else:
# PULL VALUE OUT OF THE stats AGGREGATE
s.pull = jx_expression_to_function(aggregates[s.aggregate])
nest.add(
ExprAggs(
canonical_name, {
"extended_stats": {
"script":
text(
NumberOp(s.value).partial_eval().to_es_script(
schema))
}
}, s))
def es_aggsop(es, frum, query):
query = query.copy() # WE WILL MARK UP THIS QUERY
schema = frum.schema
query_path = schema.query_path[0]
select = listwrap(query.select)
new_select = Data(
) # MAP FROM canonical_name (USED FOR NAMES IN QUERY) TO SELECT MAPPING
formula = []
for s in select:
if is_op(s.value, Variable_):
s.query_path = query_path
if s.aggregate == "count":
new_select["count_" + literal_field(s.value.var)] += [s]
else:
new_select[literal_field(s.value.var)] += [s]
elif s.aggregate:
split_select = split_expression_by_path(s.value,
schema,
lang=Painless)
for si_key, si_value in split_select.items():
if si_value:
if s.query_path:
Log.error(
"can not handle more than one depth per select")
s.query_path = si_key
formula.append(s)
acc = Aggs()
for _, many in new_select.items():
for s in many:
canonical_name = s.name
if s.aggregate in ("value_count", "count"):
columns = frum.schema.values(s.value.var,
exclude_type=(OBJECT, NESTED))
else:
columns = frum.schema.values(s.value.var)
if s.aggregate == "count":
canonical_names = []
for column in columns:
es_name = column.es_column + "_count"
if column.jx_type == EXISTS:
if column.nested_path[0] == query_path:
canonical_names.append("doc_count")
acc.add(
NestedAggs(column.nested_path[0]).add(
CountAggs(s)))
else:
canonical_names.append("value")
acc.add(
NestedAggs(column.nested_path[0]).add(
ExprAggs(es_name, {
"value_count": {
"field": column.es_column
}
}, s)))
if len(canonical_names) == 1:
s.pull = jx_expression_to_function(canonical_names[0])
else:
s.pull = jx_expression_to_function(
{"add": canonical_names})
elif s.aggregate == "median":
columns = [
c for c in columns if c.jx_type in (NUMBER, INTEGER)
]
if len(columns) != 1:
Log.error(
"Do not know how to perform median on columns with more than one type (script probably)"
)
# ES USES DIFFERENT METHOD FOR PERCENTILES
key = canonical_name + " percentile"
acc.add(
ExprAggs(
key, {
"percentiles": {
"field": first(columns).es_column,
"percents": [50]
}
}, s))
s.pull = jx_expression_to_function("values.50\\.0")
elif s.aggregate == "percentile":
columns = [
c for c in columns if c.jx_type in (NUMBER, INTEGER)
]
if len(columns) != 1:
Log.error(
"Do not know how to perform percentile on columns with more than one type (script probably)"
)
# ES USES DIFFERENT METHOD FOR PERCENTILES
key = canonical_name + " percentile"
if is_text(
s.percentile) or s.percetile < 0 or 1 < s.percentile:
Log.error(
"Expecting percentile to be a float from 0.0 to 1.0")
percent = mo_math.round(s.percentile * 100, decimal=6)
acc.add(
ExprAggs(
key, {
"percentiles": {
"field": first(columns).es_column,
"percents": [percent],
"tdigest": {
"compression": 2
}
}
}, s))
s.pull = jx_expression_to_function(
join_field(["values", text_type(percent)]))
elif s.aggregate == "cardinality":
for column in columns:
path = column.es_column + "_cardinality"
acc.add(
ExprAggs(path,
{"cardinality": {
"field": column.es_column
}}, s))
s.pull = jx_expression_to_function("value")
elif s.aggregate == "stats":
columns = [
c for c in columns if c.jx_type in (NUMBER, INTEGER)
]
if len(columns) != 1:
Log.error(
"Do not know how to perform stats on columns with more than one type (script probably)"
)
# REGULAR STATS
acc.add(
ExprAggs(canonical_name, {
"extended_stats": {
"field": first(columns).es_column
}
}, s))
s.pull = get_pull_stats()
# GET MEDIAN TOO!
select_median = s.copy()
select_median.pull = jx_expression_to_function(
{"select": [{
"name": "median",
"value": "values.50\\.0"
}]})
acc.add(
ExprAggs(
canonical_name + "_percentile", {
"percentiles": {
"field": first(columns).es_column,
"percents": [50]
}
}, select_median))
elif s.aggregate == "union":
for column in columns:
script = {
"scripted_metric": {
'init_script':
'params._agg.terms = new HashSet()',
'map_script':
'for (v in doc[' + quote(column.es_column) +
'].values) params._agg.terms.add(v);',
'combine_script':
'return params._agg.terms.toArray()',
'reduce_script':
'HashSet output = new HashSet(); for (a in params._aggs) { if (a!=null) for (v in a) {output.add(v)} } return output.toArray()',
}
}
stats_name = column.es_column
acc.add(
NestedAggs(column.nested_path[0]).add(
ExprAggs(stats_name, script, s)))
s.pull = jx_expression_to_function("value")
elif s.aggregate == "count_values":
# RETURN MAP FROM VALUE TO THE NUMBER OF TIMES FOUND IN THE DOCUMENTS
# NOT A NESTED DOC, RATHER A MULTIVALUE FIELD
for column in columns:
script = {
"scripted_metric": {
'params': {
"_agg": {}
},
'init_script':
'params._agg.terms = new HashMap()',
'map_script':
'for (v in doc[' + quote(column.es_column) +
'].values) params._agg.terms.put(v, Optional.ofNullable(params._agg.terms.get(v)).orElse(0)+1);',
'combine_script':
'return params._agg.terms',
'reduce_script':
'''
HashMap output = new HashMap();
for (agg in params._aggs) {
if (agg!=null){
for (e in agg.entrySet()) {
String key = String.valueOf(e.getKey());
output.put(key, e.getValue() + Optional.ofNullable(output.get(key)).orElse(0));
}
}
}
return output;
'''
}
}
stats_name = encode_property(column.es_column)
acc.add(
NestedAggs(column.nested_path[0]).add(
ExprAggs(stats_name, script, s)))
s.pull = jx_expression_to_function("value")
else:
if not columns:
s.pull = jx_expression_to_function(NULL)
else:
for c in columns:
acc.add(
NestedAggs(c.nested_path[0]).add(
ExprAggs(
canonical_name,
{"extended_stats": {
"field": c.es_column
}}, s)))
s.pull = jx_expression_to_function(aggregates[s.aggregate])
for i, s in enumerate(formula):
s_path = [
k for k, v in split_expression_by_path(
s.value, schema=schema, lang=Painless).items() if v
]
if len(s_path) == 0:
# FOR CONSTANTS
nest = NestedAggs(query_path)
acc.add(nest)
elif len(s_path) == 1:
nest = NestedAggs(first(s_path))
acc.add(nest)
else:
Log.error("do not know how to handle")
canonical_name = s.name
if is_op(s.value, TupleOp):
if s.aggregate == "count":
# TUPLES ALWAYS EXIST, SO COUNTING THEM IS EASY
s.pull = jx_expression_to_function("doc_count")
elif s.aggregate in ('max', 'maximum', 'min', 'minimum'):
if s.aggregate in ('max', 'maximum'):
dir = 1
op = "max"
else:
dir = -1
op = 'min'
nully = Painless[TupleOp(
[NULL] *
len(s.value.terms))].partial_eval().to_es_script(schema)
selfy = text_type(
Painless[s.value].partial_eval().to_es_script(schema))
script = {
"scripted_metric": {
'init_script':
'params._agg.best = ' + nully + ';',
'map_script':
'params._agg.best = ' + expand_template(
MAX_OF_TUPLE, {
"expr1": "params._agg.best",
"expr2": selfy,
"dir": dir,
"op": op
}) + ";",
'combine_script':
'return params._agg.best',
'reduce_script':
'return params._aggs.stream().' + op + '(' +
expand_template(COMPARE_TUPLE, {
"dir": dir,
"op": op
}) + ').get()',
}
}
nest.add(
NestedAggs(query_path).add(
ExprAggs(canonical_name, script, s)))
s.pull = jx_expression_to_function("value")
else:
Log.error("{{agg}} is not a supported aggregate over a tuple",
agg=s.aggregate)
elif s.aggregate == "count":
nest.add(
ExprAggs(
canonical_name, {
"value_count": {
"script":
text_type(Painless[
s.value].partial_eval().to_es_script(schema))
}
}, s))
s.pull = jx_expression_to_function("value")
elif s.aggregate == "median":
# ES USES DIFFERENT METHOD FOR PERCENTILES THAN FOR STATS AND COUNT
key = literal_field(canonical_name + " percentile")
nest.add(
ExprAggs(
key, {
"percentiles": {
"script":
text_type(Painless[s.value].to_es_script(schema)),
"percents": [50]
}
}, s))
s.pull = jx_expression_to_function(join_field(["50.0"]))
elif s.aggregate == "percentile":
# ES USES DIFFERENT METHOD FOR PERCENTILES THAN FOR STATS AND COUNT
key = literal_field(canonical_name + " percentile")
percent = mo_math.round(s.percentile * 100, decimal=6)
nest.add(
ExprAggs(
key, {
"percentiles": {
"script":
text_type(Painless[s.value].to_es_script(schema)),
"percents": [percent]
}
}, s))
s.pull = jx_expression_to_function(
join_field(["values", text_type(percent)]))
elif s.aggregate == "cardinality":
# ES USES DIFFERENT METHOD FOR CARDINALITY
key = canonical_name + " cardinality"
nest.add(
ExprAggs(
key, {
"cardinality": {
"script":
text_type(Painless[s.value].to_es_script(schema))
}
}, s))
s.pull = jx_expression_to_function("value")
elif s.aggregate == "stats":
# REGULAR STATS
nest.add(
ExprAggs(
canonical_name, {
"extended_stats": {
"script":
text_type(Painless[s.value].to_es_script(schema))
}
}, s))
s.pull = get_pull_stats()
# GET MEDIAN TOO!
select_median = s.copy()
select_median.pull = jx_expression_to_function(
{"select": [{
"name": "median",
"value": "values.50\\.0"
}]})
nest.add(
ExprAggs(
canonical_name + "_percentile", {
"percentiles": {
"script":
text_type(Painless[s.value].to_es_script(schema)),
"percents": [50]
}
}, select_median))
s.pull = get_pull_stats()
elif s.aggregate == "union":
# USE TERMS AGGREGATE TO SIMULATE union
nest.add(
TermsAggs(
canonical_name, {
"script_field":
text_type(Painless[s.value].to_es_script(schema))
}, s))
s.pull = jx_expression_to_function("key")
else:
# PULL VALUE OUT OF THE stats AGGREGATE
s.pull = jx_expression_to_function(aggregates[s.aggregate])
nest.add(
ExprAggs(
canonical_name, {
"extended_stats": {
"script":
text_type(
NumberOp(s.value).partial_eval().to_es_script(
schema))
}
}, s))
acc = NestedAggs(query_path).add(acc)
split_decoders = get_decoders_by_path(query)
split_wheres = split_expression_by_path(query.where,
schema=frum.schema,
lang=ES52)
start = 0
decoders = [None] * (len(query.edges) + len(query.groupby))
paths = list(reversed(sorted(split_wheres.keys() | split_decoders.keys())))
for path in paths:
literal_path = literal_field(path)
decoder = split_decoders[literal_path]
where = split_wheres[literal_path]
for d in decoder:
decoders[d.edge.dim] = d
acc = d.append_query(path, acc)
start += d.num_columns
if where:
acc = FilterAggs("_filter", AndOp(where), None).add(acc)
acc = NestedAggs(path).add(acc)
acc = NestedAggs('.').add(acc)
acc = simplify(acc)
es_query = wrap(acc.to_es(schema))
es_query.size = 0
with Timer("ES query time", silent=not DEBUG) as es_duration:
result = es_post(es, es_query, query.limit)
try:
format_time = Timer("formatting", silent=not DEBUG)
with format_time:
# result.aggregations.doc_count = coalesce(result.aggregations.doc_count, result.hits.total) # IT APPEARS THE OLD doc_count IS GONE
aggs = unwrap(result.aggregations)
formatter, groupby_formatter, aggop_formatter, mime_type = format_dispatch[
query.format]
if query.edges:
output = formatter(aggs, acc, query, decoders, select)
elif query.groupby:
output = groupby_formatter(aggs, acc, query, decoders, select)
else:
output = aggop_formatter(aggs, acc, query, decoders, select)
output.meta.timing.formatting = format_time.duration
output.meta.timing.es_search = es_duration.duration
output.meta.content_type = mime_type
output.meta.es_query = es_query
return output
except Exception as e:
if query.format not in format_dispatch:
Log.error("Format {{format|quote}} not supported yet",
format=query.format,
cause=e)
Log.error("Some problem", cause=e)
def es_setop(es, query):
schema = query.frum.schema
query_path = schema.query_path[0]
split_select = {".": ESSelect('.')}
def get_select(path):
es_select = split_select.get(path)
if not es_select:
es_select = split_select[path] = ESSelect(path)
return es_select
selects = wrap([unwrap(s.copy()) for s in listwrap(query.select)])
new_select = FlatList()
put_index = 0
for select in selects:
# IF THERE IS A *, THEN INSERT THE EXTRA COLUMNS
if is_op(select.value, LeavesOp) and is_op(select.value.term,
Variable):
term = select.value.term
leaves = schema.leaves(term.var)
for c in leaves:
full_name = concat_field(
select.name, relative_field(untype_path(c.name), term.var))
if c.jx_type == NESTED:
get_select('.').use_source = True
new_select.append({
"name": full_name,
"value": Variable(c.es_column),
"put": {
"name": literal_field(full_name),
"index": put_index,
"child": "."
},
"pull": get_pull_source(c.es_column)
})
put_index += 1
else:
get_select(c.nested_path[0]).fields.append(c.es_column)
new_select.append({
"name": full_name,
"value": Variable(c.es_column),
"put": {
"name": literal_field(full_name),
"index": put_index,
"child": "."
}
})
put_index += 1
elif is_op(select.value, Variable):
s_column = select.value.var
if s_column == ".":
# PULL ALL SOURCE
get_select('.').use_source = True
new_select.append({
"name": select.name,
"value": select.value,
"put": {
"name": select.name,
"index": put_index,
"child": "."
},
"pull": get_pull_source(".")
})
continue
leaves = schema.leaves(s_column) # LEAVES OF OBJECT
# nested_selects = {}
if leaves:
if any(c.jx_type == NESTED for c in leaves):
# PULL WHOLE NESTED ARRAYS
get_select('.').use_source = True
for c in leaves:
if len(
c.nested_path
) == 1: # NESTED PROPERTIES ARE IGNORED, CAPTURED BY THESE FIRST LEVEL PROPERTIES
pre_child = join_field(
decode_property(n)
for n in split_field(c.name))
new_select.append({
"name":
select.name,
"value":
Variable(c.es_column),
"put": {
"name":
select.name,
"index":
put_index,
"child":
untype_path(
relative_field(pre_child, s_column))
},
"pull":
get_pull_source(c.es_column)
})
else:
# PULL ONLY WHAT'S NEEDED
for c in leaves:
c_nested_path = c.nested_path[0]
if c_nested_path == ".":
if c.es_column == "_id":
new_select.append({
"name":
select.name,
"value":
Variable(c.es_column),
"put": {
"name": select.name,
"index": put_index,
"child": "."
},
"pull":
lambda row: row._id
})
elif c.jx_type == NESTED:
get_select('.').use_source = True
pre_child = join_field(
decode_property(n)
for n in split_field(c.name))
new_select.append({
"name":
select.name,
"value":
Variable(c.es_column),
"put": {
"name":
select.name,
"index":
put_index,
"child":
untype_path(
relative_field(
pre_child, s_column))
},
"pull":
get_pull_source(c.es_column)
})
else:
get_select(c_nested_path).fields.append(
c.es_column)
pre_child = join_field(
decode_property(n)
for n in split_field(c.name))
new_select.append({
"name":
select.name,
"value":
Variable(c.es_column),
"put": {
"name":
select.name,
"index":
put_index,
"child":
untype_path(
relative_field(
pre_child, s_column))
}
})
else:
es_select = get_select(c_nested_path)
es_select.fields.append(c.es_column)
child = relative_field(
untype_path(
relative_field(c.name,
schema.query_path[0])),
s_column)
pull = accumulate_nested_doc(
c_nested_path,
Variable(
relative_field(
s_column, unnest_path(c_nested_path))))
new_select.append({
"name": select.name,
"value": select.value,
"put": {
"name": select.name,
"index": put_index,
"child": child
},
"pull": pull
})
else:
new_select.append({
"name": select.name,
"value": Variable("$dummy"),
"put": {
"name": select.name,
"index": put_index,
"child": "."
}
})
put_index += 1
else:
split_scripts = split_expression_by_path(select.value,
schema,
lang=Painless)
for p, script in split_scripts.items():
es_select = get_select(p)
es_select.scripts[select.name] = {
"script":
text_type(Painless[first(
script)].partial_eval().to_es_script(schema))
}
new_select.append({
"name":
select.name,
"pull":
jx_expression_to_function("fields." +
literal_field(select.name)),
"put": {
"name": select.name,
"index": put_index,
"child": "."
}
})
put_index += 1
for n in new_select:
if n.pull:
continue
elif is_op(n.value, Variable):
if get_select('.').use_source:
n.pull = get_pull_source(n.value.var)
elif n.value == "_id":
n.pull = jx_expression_to_function("_id")
else:
n.pull = jx_expression_to_function(
concat_field("fields", literal_field(n.value.var)))
else:
Log.error("Do not know what to do")
split_wheres = split_expression_by_path(query.where, schema, lang=ES52)
es_query = es_query_proto(query_path, split_select, split_wheres, schema)
es_query.size = coalesce(query.limit, DEFAULT_LIMIT)
es_query.sort = jx_sort_to_es_sort(query.sort, schema)
with Timer("call to ES", silent=DEBUG) as call_timer:
data = es_post(es, es_query, query.limit)
T = data.hits.hits
# Log.note("{{output}}", output=T)
try:
formatter, groupby_formatter, mime_type = format_dispatch[query.format]
with Timer("formatter", silent=True):
output = formatter(T, new_select, query)
output.meta.timing.es = call_timer.duration
output.meta.content_type = mime_type
output.meta.es_query = es_query
return output
except Exception as e:
Log.error("problem formatting", e)
def get_selects(query):
schema = query.frum.schema
query_level = len(schema.query_path)
query_path = schema.query_path[0]
# SPLIT select INTO ES_SELECT AND RESULTSET SELECT
split_select = OrderedDict((p, ESSelectOp(p)) for p in schema.query_path)
def expand_split_select(c_nested_path):
es_select = split_select.get(c_nested_path)
if not es_select:
temp = [(k, v) for k, v in split_select.items()]
split_select.clear()
split_select.update({c_nested_path: ESSelectOp(c_nested_path)})
split_select.update(temp)
return split_select[c_nested_path]
new_select = FlatList()
post_expressions = {}
selects = list_to_data([unwrap(s.copy()) for s in listwrap(query.select)])
# WHAT PATH IS _source USED, IF ANY?
for select in selects:
# IF THERE IS A *, THEN INSERT THE EXTRA COLUMNS
if is_op(select.value, LeavesOp) and is_op(select.value.term,
Variable):
term = select.value.term
leaves = schema.leaves(term.var)
if any(c.jx_type == NESTED for c in leaves):
split_select["."].source_path = "."
elif is_op(select.value, Variable):
for selected_column in schema.values(select.value.var,
exclude_type=(OBJECT,
EXISTS)):
if selected_column.jx_type == NESTED:
expand_split_select(
selected_column.es_column
).source_path = selected_column.es_column
continue
leaves = schema.leaves(selected_column.es_column)
for c in leaves:
if c.jx_type == NESTED:
split_select[c.es_column].source_path = c.es_column
# IF WE GET THE SOURCE FOR PARENT, WE ASSUME WE GOT SOURCE FOR CHILD
source_path = None
source_level = 0
for level, es_select in enumerate(reversed(list(split_select.values()))):
if source_path:
es_select.source_path = source_path
elif es_select.source_path:
source_level = level + 1
source_path = es_select.source_path
def get_pull_source(c):
nested_path = c.nested_path
nested_level = len(nested_path)
pos = text(nested_level)
if nested_level <= query_level:
if not source_level or nested_level < source_level:
field = join_field([pos, "fields", c.es_column])
return jx_expression_to_function(field)
elif nested_level == source_level:
field = relative_field(c.es_column, nested_path[0])
def pull_source(row):
return untyped(row.get(pos, Null)._source[field])
return pull_source
else:
field = relative_field(c.es_column, nested_path[0])
def pull_property(row):
return untyped(row.get(pos, Null)[field])
return pull_property
else:
pos = text(query_level)
if not source_level or nested_level < source_level:
# PULL FIELDS AND THEN AGGREGATE THEM
value = jx_expression_to_function(
join_field(["fields", c.es_column]))
name = literal_field(nested_path[0])
index = jx_expression_to_function("_nested.offset")
def pull_nested_field(doc):
hits = doc.get(pos, Null).inner_hits[name].hits.hits
if not hits:
return []
temp = [(index(h), value(h)) for h in hits]
acc = [None] * len(temp)
for i, v in temp:
acc[i] = unwraplist(v)
return acc
return pull_nested_field
else:
# PULL SOURCES
value = jx_expression_to_function(
concat_field("_source",
relative_field(c.es_column, nested_path[0])))
name = literal_field(nested_path[0])
index = jx_expression_to_function(
join_field(["_nested"] * (len(c.nested_path) - 1) +
["offset"]))
def pull_nested_source(doc):
hits = doc.get(pos, Null).inner_hits[name].hits.hits
if not hits:
return []
temp = [(index(h), value(h)) for h in hits]
acc = [None] * len(temp)
for i, v in temp:
acc[i] = untyped(v)
return acc
return pull_nested_source
put_index = 0
for select in selects:
# IF THERE IS A *, THEN INSERT THE EXTRA COLUMNS
if is_op(select.value, LeavesOp) and is_op(select.value.term,
Variable):
term = select.value.term
leaves = schema.leaves(term.var)
for c in leaves:
c_nested_path = c.nested_path[0]
simple_name = relative_field(c.es_column,
query_path).lstrip(".")
name = concat_field(select.name, untype_path(simple_name))
put_name = concat_field(
select.name, literal_field(untype_path(simple_name)))
split_select[c_nested_path].fields.append(c.es_column)
new_select.append({
"name": name,
"value": Variable(c.es_column),
"put": {
"name": put_name,
"index": put_index,
"child": ".",
},
"pull": get_pull_source(c),
})
put_index += 1
elif is_op(select.value, Variable):
if select.value.var == ".":
# PULL ALL SOURCE
new_select.append({
"name":
select.name,
"value":
select.value,
"put": {
"name": select.name,
"index": put_index,
"child": "."
},
"pull":
get_pull_source(
Data(es_column=query_path,
nested_path=schema.query_path)),
})
continue
for selected_column in schema.values(select.value.var,
exclude_type=(EXISTS,
OBJECT)):
if selected_column.jx_type == NESTED:
new_select.append({
"name":
select.name,
"value":
select.value,
"put": {
"name": select.name,
"index": put_index,
"child": "."
},
"pull":
get_pull_source(
Data(
es_column=selected_column.es_column,
nested_path=(selected_column.es_column, ) +
selected_column.nested_path,
)),
})
continue
leaves = schema.leaves(
selected_column.es_column,
exclude_type=INTERNAL) # LEAVES OF OBJECT
if leaves:
for c in leaves:
if c.es_column == "_id":
new_select.append({
"name": select.name,
"value": Variable(c.es_column),
"put": {
"name": select.name,
"index": put_index,
"child": ".",
},
"pull": pull_id,
})
continue
c_nested_path = c.nested_path[0]
expand_split_select(c_nested_path).fields.append(
c.es_column)
child = untype_path(
relative_field(
c.es_column,
selected_column.es_column,
))
new_select.append({
"name": select.name,
"value": Variable(c.es_column),
"put": {
"name": select.name,
"index": put_index,
"child": child,
},
"pull": get_pull_source(c),
})
else:
new_select.append({
"name": select.name,
"value": NULL,
"put": {
"name": select.name,
"index": put_index,
"child": "."
},
})
put_index += 1
else:
op, split_scripts = split_expression_by_path(select.value,
schema,
lang=Painless)
for p, script in split_scripts.items():
es_select = split_select[p]
es_select.scripts[select.name] = {
"script":
text(Painless[script].partial_eval().to_es_script(schema))
}
new_select.append({
"name":
select.name,
"pull":
jx_expression_to_function(
join_field([
text(p),
"fields",
select.name,
])),
"put": {
"name": select.name,
"index": put_index,
"child": "."
},
})
put_index += 1
def inners(query_path, parent_pos):
"""
:param query_path:
:return: ITERATOR OVER TUPLES ROWS AS TUPLES, WHERE row[len(nested_path)] HAS INNER HITS
AND row[0] HAS post_expressions
"""
pos = text(int(parent_pos) + 1)
if not query_path:
def base_case(row):
extra = {}
for k, e in post_expressions.items():
extra[k] = e(row)
row["0"] = extra
yield row
return base_case
if pos == "1":
more = inners(query_path[:-1], "1")
def first_case(results):
for result in results:
for hit in result.hits.hits:
seed = {"0": None, pos: hit}
for row in more(seed):
yield row
return first_case
else:
more = inners(query_path[:-1], pos)
if source_path and source_path < query_path[-1]:
rel_path = relative_field(query_path[-1], source_path)
def source(acc):
for inner_row in acc[parent_pos][rel_path]:
acc[pos] = inner_row
for tt in more(acc):
yield tt
return source
else:
path = literal_field(query_path[-1])
def recurse(acc):
hits = acc[parent_pos].inner_hits[path].hits.hits
if hits:
for inner_row in hits:
acc[pos] = inner_row
for tt in more(acc):
yield tt
else:
for tt in more(acc):
yield tt
return recurse
return new_select, split_select, inners(schema.query_path, "0")