terraform/helper/schema/field_reader_config.go

293 lines
7.2 KiB
Go
Raw Normal View History

package schema
import (
"fmt"
2015-01-09 03:02:19 +01:00
"strconv"
"strings"
2015-01-09 03:02:19 +01:00
"sync"
"github.com/hashicorp/terraform/terraform"
"github.com/mitchellh/mapstructure"
)
// ConfigFieldReader reads fields out of an untyped map[string]string to the
// best of its ability. It also applies defaults from the Schema. (The other
// field readers do not need default handling because they source fully
// populated data structures.)
type ConfigFieldReader struct {
Config *terraform.ResourceConfig
Schema map[string]*Schema
2015-01-09 03:02:19 +01:00
Change Set internals and make (extreme) performance improvements Changing the Set internals makes a lot of sense as it saves doing conversions in multiple places and gives a central place to alter the key when a item is computed. This will have no side effects other then that the ordering is now based on strings instead on integers, so the order will be different. This will however have no effect on existing configs as these will use the individual codes/keys and not the ordering to determine if there is a diff or not. Lastly (but I think also most importantly) there is a fix in this PR that makes diffing sets extremely more performand. Before a full diff required reading the complete Set for every single parameter/attribute you wanted to diff, while now it only gets that specific parameter. We have a use case where we have a Set that has 18 parameters and the set consist of about 600 items (don't ask :wink:). So when doing a diff it would take 100% CPU of all cores and stay that way for almost an hour before being able to complete the diff. Debugging this we learned that for retrieving every single parameter it made over 52.000 calls to `func (c *ResourceConfig) get(..)`. In this function a slice is created and used only for the duration of the call, so the time needed to create all needed slices and on the other hand the time the garbage collector needed to clean them up again caused the system to cripple itself. Next to that there are also some expensive reflect calls in this function which also claimed a fair amount of CPU time. After this fix the number of calls needed to get a single parameter dropped from 52.000+ to only 2! :smiley:
2015-11-18 11:24:04 +01:00
indexMaps map[string]map[string]int
once sync.Once
}
func (r *ConfigFieldReader) ReadField(address []string) (FieldReadResult, error) {
Change Set internals and make (extreme) performance improvements Changing the Set internals makes a lot of sense as it saves doing conversions in multiple places and gives a central place to alter the key when a item is computed. This will have no side effects other then that the ordering is now based on strings instead on integers, so the order will be different. This will however have no effect on existing configs as these will use the individual codes/keys and not the ordering to determine if there is a diff or not. Lastly (but I think also most importantly) there is a fix in this PR that makes diffing sets extremely more performand. Before a full diff required reading the complete Set for every single parameter/attribute you wanted to diff, while now it only gets that specific parameter. We have a use case where we have a Set that has 18 parameters and the set consist of about 600 items (don't ask :wink:). So when doing a diff it would take 100% CPU of all cores and stay that way for almost an hour before being able to complete the diff. Debugging this we learned that for retrieving every single parameter it made over 52.000 calls to `func (c *ResourceConfig) get(..)`. In this function a slice is created and used only for the duration of the call, so the time needed to create all needed slices and on the other hand the time the garbage collector needed to clean them up again caused the system to cripple itself. Next to that there are also some expensive reflect calls in this function which also claimed a fair amount of CPU time. After this fix the number of calls needed to get a single parameter dropped from 52.000+ to only 2! :smiley:
2015-11-18 11:24:04 +01:00
r.once.Do(func() { r.indexMaps = make(map[string]map[string]int) })
2015-01-09 03:02:19 +01:00
return r.readField(address, false)
}
func (r *ConfigFieldReader) readField(
address []string, nested bool) (FieldReadResult, error) {
schemaList := addrToSchema(address, r.Schema)
if len(schemaList) == 0 {
return FieldReadResult{}, nil
}
2015-01-09 03:02:19 +01:00
if !nested {
// If we have a set anywhere in the address, then we need to
// read that set out in order and actually replace that part of
// the address with the real list index. i.e. set.50 might actually
// map to set.12 in the config, since it is in list order in the
// config, not indexed by set value.
for i, v := range schemaList {
// Sets are the only thing that cause this issue.
if v.Type != TypeSet {
continue
}
// If we're at the end of the list, then we don't have to worry
// about this because we're just requesting the whole set.
if i == len(schemaList)-1 {
continue
}
// If we're looking for the count, then ignore...
if address[i+1] == "#" {
continue
}
Change Set internals and make (extreme) performance improvements Changing the Set internals makes a lot of sense as it saves doing conversions in multiple places and gives a central place to alter the key when a item is computed. This will have no side effects other then that the ordering is now based on strings instead on integers, so the order will be different. This will however have no effect on existing configs as these will use the individual codes/keys and not the ordering to determine if there is a diff or not. Lastly (but I think also most importantly) there is a fix in this PR that makes diffing sets extremely more performand. Before a full diff required reading the complete Set for every single parameter/attribute you wanted to diff, while now it only gets that specific parameter. We have a use case where we have a Set that has 18 parameters and the set consist of about 600 items (don't ask :wink:). So when doing a diff it would take 100% CPU of all cores and stay that way for almost an hour before being able to complete the diff. Debugging this we learned that for retrieving every single parameter it made over 52.000 calls to `func (c *ResourceConfig) get(..)`. In this function a slice is created and used only for the duration of the call, so the time needed to create all needed slices and on the other hand the time the garbage collector needed to clean them up again caused the system to cripple itself. Next to that there are also some expensive reflect calls in this function which also claimed a fair amount of CPU time. After this fix the number of calls needed to get a single parameter dropped from 52.000+ to only 2! :smiley:
2015-11-18 11:24:04 +01:00
indexMap, ok := r.indexMaps[strings.Join(address[:i+1], ".")]
if !ok {
// Get the set so we can get the index map that tells us the
// mapping of the hash code to the list index
_, err := r.readSet(address[:i+1], v)
if err != nil {
return FieldReadResult{}, err
}
indexMap = r.indexMaps[strings.Join(address[:i+1], ".")]
2015-01-09 03:02:19 +01:00
}
Change Set internals and make (extreme) performance improvements Changing the Set internals makes a lot of sense as it saves doing conversions in multiple places and gives a central place to alter the key when a item is computed. This will have no side effects other then that the ordering is now based on strings instead on integers, so the order will be different. This will however have no effect on existing configs as these will use the individual codes/keys and not the ordering to determine if there is a diff or not. Lastly (but I think also most importantly) there is a fix in this PR that makes diffing sets extremely more performand. Before a full diff required reading the complete Set for every single parameter/attribute you wanted to diff, while now it only gets that specific parameter. We have a use case where we have a Set that has 18 parameters and the set consist of about 600 items (don't ask :wink:). So when doing a diff it would take 100% CPU of all cores and stay that way for almost an hour before being able to complete the diff. Debugging this we learned that for retrieving every single parameter it made over 52.000 calls to `func (c *ResourceConfig) get(..)`. In this function a slice is created and used only for the duration of the call, so the time needed to create all needed slices and on the other hand the time the garbage collector needed to clean them up again caused the system to cripple itself. Next to that there are also some expensive reflect calls in this function which also claimed a fair amount of CPU time. After this fix the number of calls needed to get a single parameter dropped from 52.000+ to only 2! :smiley:
2015-11-18 11:24:04 +01:00
index, ok := indexMap[address[i+1]]
2015-01-09 03:02:19 +01:00
if !ok {
return FieldReadResult{}, nil
}
address[i+1] = strconv.FormatInt(int64(index), 10)
}
}
k := strings.Join(address, ".")
schema := schemaList[len(schemaList)-1]
switch schema.Type {
case TypeBool, TypeFloat, TypeInt, TypeString:
return r.readPrimitive(k, schema)
case TypeList:
2015-01-09 03:02:19 +01:00
return readListField(&nestedConfigFieldReader{r}, address, schema)
case TypeMap:
return r.readMap(k)
case TypeSet:
Change Set internals and make (extreme) performance improvements Changing the Set internals makes a lot of sense as it saves doing conversions in multiple places and gives a central place to alter the key when a item is computed. This will have no side effects other then that the ordering is now based on strings instead on integers, so the order will be different. This will however have no effect on existing configs as these will use the individual codes/keys and not the ordering to determine if there is a diff or not. Lastly (but I think also most importantly) there is a fix in this PR that makes diffing sets extremely more performand. Before a full diff required reading the complete Set for every single parameter/attribute you wanted to diff, while now it only gets that specific parameter. We have a use case where we have a Set that has 18 parameters and the set consist of about 600 items (don't ask :wink:). So when doing a diff it would take 100% CPU of all cores and stay that way for almost an hour before being able to complete the diff. Debugging this we learned that for retrieving every single parameter it made over 52.000 calls to `func (c *ResourceConfig) get(..)`. In this function a slice is created and used only for the duration of the call, so the time needed to create all needed slices and on the other hand the time the garbage collector needed to clean them up again caused the system to cripple itself. Next to that there are also some expensive reflect calls in this function which also claimed a fair amount of CPU time. After this fix the number of calls needed to get a single parameter dropped from 52.000+ to only 2! :smiley:
2015-11-18 11:24:04 +01:00
return r.readSet(address, schema)
case typeObject:
2015-01-09 03:02:19 +01:00
return readObjectField(
&nestedConfigFieldReader{r},
address, schema.Elem.(map[string]*Schema))
default:
2015-01-11 01:04:01 +01:00
panic(fmt.Sprintf("Unknown type: %s", schema.Type))
}
}
func (r *ConfigFieldReader) readMap(k string) (FieldReadResult, error) {
// We want both the raw value and the interpolated. We use the interpolated
// to store actual values and we use the raw one to check for
// computed keys. Actual values are obtained in the switch, depending on
// the type of the raw value.
mraw, ok := r.Config.GetRaw(k)
if !ok {
return FieldReadResult{}, nil
}
result := make(map[string]interface{})
computed := false
switch m := mraw.(type) {
case string:
// This is a map which has come out of an interpolated variable, so we
// can just get the value directly from config. Values cannot be computed
// currently.
v, _ := r.Config.Get(k)
// If this isn't a map[string]interface, it must be computed.
mapV, ok := v.(map[string]interface{})
if !ok {
return FieldReadResult{
Exists: true,
Computed: true,
}, nil
}
// Otherwise we can proceed as usual.
for i, iv := range mapV {
result[i] = iv
}
case []interface{}:
for i, innerRaw := range m {
Change Set internals and make (extreme) performance improvements Changing the Set internals makes a lot of sense as it saves doing conversions in multiple places and gives a central place to alter the key when a item is computed. This will have no side effects other then that the ordering is now based on strings instead on integers, so the order will be different. This will however have no effect on existing configs as these will use the individual codes/keys and not the ordering to determine if there is a diff or not. Lastly (but I think also most importantly) there is a fix in this PR that makes diffing sets extremely more performand. Before a full diff required reading the complete Set for every single parameter/attribute you wanted to diff, while now it only gets that specific parameter. We have a use case where we have a Set that has 18 parameters and the set consist of about 600 items (don't ask :wink:). So when doing a diff it would take 100% CPU of all cores and stay that way for almost an hour before being able to complete the diff. Debugging this we learned that for retrieving every single parameter it made over 52.000 calls to `func (c *ResourceConfig) get(..)`. In this function a slice is created and used only for the duration of the call, so the time needed to create all needed slices and on the other hand the time the garbage collector needed to clean them up again caused the system to cripple itself. Next to that there are also some expensive reflect calls in this function which also claimed a fair amount of CPU time. After this fix the number of calls needed to get a single parameter dropped from 52.000+ to only 2! :smiley:
2015-11-18 11:24:04 +01:00
for ik := range innerRaw.(map[string]interface{}) {
key := fmt.Sprintf("%s.%d.%s", k, i, ik)
if r.Config.IsComputed(key) {
computed = true
break
}
v, _ := r.Config.Get(key)
result[ik] = v
}
}
case []map[string]interface{}:
for i, innerRaw := range m {
Change Set internals and make (extreme) performance improvements Changing the Set internals makes a lot of sense as it saves doing conversions in multiple places and gives a central place to alter the key when a item is computed. This will have no side effects other then that the ordering is now based on strings instead on integers, so the order will be different. This will however have no effect on existing configs as these will use the individual codes/keys and not the ordering to determine if there is a diff or not. Lastly (but I think also most importantly) there is a fix in this PR that makes diffing sets extremely more performand. Before a full diff required reading the complete Set for every single parameter/attribute you wanted to diff, while now it only gets that specific parameter. We have a use case where we have a Set that has 18 parameters and the set consist of about 600 items (don't ask :wink:). So when doing a diff it would take 100% CPU of all cores and stay that way for almost an hour before being able to complete the diff. Debugging this we learned that for retrieving every single parameter it made over 52.000 calls to `func (c *ResourceConfig) get(..)`. In this function a slice is created and used only for the duration of the call, so the time needed to create all needed slices and on the other hand the time the garbage collector needed to clean them up again caused the system to cripple itself. Next to that there are also some expensive reflect calls in this function which also claimed a fair amount of CPU time. After this fix the number of calls needed to get a single parameter dropped from 52.000+ to only 2! :smiley:
2015-11-18 11:24:04 +01:00
for ik := range innerRaw {
key := fmt.Sprintf("%s.%d.%s", k, i, ik)
if r.Config.IsComputed(key) {
computed = true
break
}
v, _ := r.Config.Get(key)
result[ik] = v
}
}
case map[string]interface{}:
Change Set internals and make (extreme) performance improvements Changing the Set internals makes a lot of sense as it saves doing conversions in multiple places and gives a central place to alter the key when a item is computed. This will have no side effects other then that the ordering is now based on strings instead on integers, so the order will be different. This will however have no effect on existing configs as these will use the individual codes/keys and not the ordering to determine if there is a diff or not. Lastly (but I think also most importantly) there is a fix in this PR that makes diffing sets extremely more performand. Before a full diff required reading the complete Set for every single parameter/attribute you wanted to diff, while now it only gets that specific parameter. We have a use case where we have a Set that has 18 parameters and the set consist of about 600 items (don't ask :wink:). So when doing a diff it would take 100% CPU of all cores and stay that way for almost an hour before being able to complete the diff. Debugging this we learned that for retrieving every single parameter it made over 52.000 calls to `func (c *ResourceConfig) get(..)`. In this function a slice is created and used only for the duration of the call, so the time needed to create all needed slices and on the other hand the time the garbage collector needed to clean them up again caused the system to cripple itself. Next to that there are also some expensive reflect calls in this function which also claimed a fair amount of CPU time. After this fix the number of calls needed to get a single parameter dropped from 52.000+ to only 2! :smiley:
2015-11-18 11:24:04 +01:00
for ik := range m {
key := fmt.Sprintf("%s.%s", k, ik)
if r.Config.IsComputed(key) {
computed = true
break
}
v, _ := r.Config.Get(key)
result[ik] = v
}
default:
panic(fmt.Sprintf("unknown type: %#v", mraw))
}
var value interface{}
if !computed {
value = result
}
return FieldReadResult{
Value: value,
Exists: true,
Computed: computed,
}, nil
}
func (r *ConfigFieldReader) readPrimitive(
k string, schema *Schema) (FieldReadResult, error) {
raw, ok := r.Config.Get(k)
if !ok {
// Nothing in config, but we might still have a default from the schema
var err error
raw, err = schema.DefaultValue()
if err != nil {
return FieldReadResult{}, fmt.Errorf("%s, error loading default: %s", k, err)
}
if raw == nil {
return FieldReadResult{}, nil
}
}
var result string
if err := mapstructure.WeakDecode(raw, &result); err != nil {
return FieldReadResult{}, err
}
computed := r.Config.IsComputed(k)
returnVal, err := stringToPrimitive(result, computed, schema)
if err != nil {
return FieldReadResult{}, err
}
return FieldReadResult{
Value: returnVal,
Exists: true,
Computed: computed,
}, nil
}
func (r *ConfigFieldReader) readSet(
Change Set internals and make (extreme) performance improvements Changing the Set internals makes a lot of sense as it saves doing conversions in multiple places and gives a central place to alter the key when a item is computed. This will have no side effects other then that the ordering is now based on strings instead on integers, so the order will be different. This will however have no effect on existing configs as these will use the individual codes/keys and not the ordering to determine if there is a diff or not. Lastly (but I think also most importantly) there is a fix in this PR that makes diffing sets extremely more performand. Before a full diff required reading the complete Set for every single parameter/attribute you wanted to diff, while now it only gets that specific parameter. We have a use case where we have a Set that has 18 parameters and the set consist of about 600 items (don't ask :wink:). So when doing a diff it would take 100% CPU of all cores and stay that way for almost an hour before being able to complete the diff. Debugging this we learned that for retrieving every single parameter it made over 52.000 calls to `func (c *ResourceConfig) get(..)`. In this function a slice is created and used only for the duration of the call, so the time needed to create all needed slices and on the other hand the time the garbage collector needed to clean them up again caused the system to cripple itself. Next to that there are also some expensive reflect calls in this function which also claimed a fair amount of CPU time. After this fix the number of calls needed to get a single parameter dropped from 52.000+ to only 2! :smiley:
2015-11-18 11:24:04 +01:00
address []string, schema *Schema) (FieldReadResult, error) {
indexMap := make(map[string]int)
// Create the set that will be our result
set := schema.ZeroValue().(*Set)
2015-01-09 03:02:19 +01:00
raw, err := readListField(&nestedConfigFieldReader{r}, address, schema)
if err != nil {
Change Set internals and make (extreme) performance improvements Changing the Set internals makes a lot of sense as it saves doing conversions in multiple places and gives a central place to alter the key when a item is computed. This will have no side effects other then that the ordering is now based on strings instead on integers, so the order will be different. This will however have no effect on existing configs as these will use the individual codes/keys and not the ordering to determine if there is a diff or not. Lastly (but I think also most importantly) there is a fix in this PR that makes diffing sets extremely more performand. Before a full diff required reading the complete Set for every single parameter/attribute you wanted to diff, while now it only gets that specific parameter. We have a use case where we have a Set that has 18 parameters and the set consist of about 600 items (don't ask :wink:). So when doing a diff it would take 100% CPU of all cores and stay that way for almost an hour before being able to complete the diff. Debugging this we learned that for retrieving every single parameter it made over 52.000 calls to `func (c *ResourceConfig) get(..)`. In this function a slice is created and used only for the duration of the call, so the time needed to create all needed slices and on the other hand the time the garbage collector needed to clean them up again caused the system to cripple itself. Next to that there are also some expensive reflect calls in this function which also claimed a fair amount of CPU time. After this fix the number of calls needed to get a single parameter dropped from 52.000+ to only 2! :smiley:
2015-11-18 11:24:04 +01:00
return FieldReadResult{}, err
}
if !raw.Exists {
Change Set internals and make (extreme) performance improvements Changing the Set internals makes a lot of sense as it saves doing conversions in multiple places and gives a central place to alter the key when a item is computed. This will have no side effects other then that the ordering is now based on strings instead on integers, so the order will be different. This will however have no effect on existing configs as these will use the individual codes/keys and not the ordering to determine if there is a diff or not. Lastly (but I think also most importantly) there is a fix in this PR that makes diffing sets extremely more performand. Before a full diff required reading the complete Set for every single parameter/attribute you wanted to diff, while now it only gets that specific parameter. We have a use case where we have a Set that has 18 parameters and the set consist of about 600 items (don't ask :wink:). So when doing a diff it would take 100% CPU of all cores and stay that way for almost an hour before being able to complete the diff. Debugging this we learned that for retrieving every single parameter it made over 52.000 calls to `func (c *ResourceConfig) get(..)`. In this function a slice is created and used only for the duration of the call, so the time needed to create all needed slices and on the other hand the time the garbage collector needed to clean them up again caused the system to cripple itself. Next to that there are also some expensive reflect calls in this function which also claimed a fair amount of CPU time. After this fix the number of calls needed to get a single parameter dropped from 52.000+ to only 2! :smiley:
2015-11-18 11:24:04 +01:00
return FieldReadResult{Value: set}, nil
}
// If the list is computed, the set is necessarilly computed
if raw.Computed {
return FieldReadResult{
Value: set,
Exists: true,
Computed: raw.Computed,
Change Set internals and make (extreme) performance improvements Changing the Set internals makes a lot of sense as it saves doing conversions in multiple places and gives a central place to alter the key when a item is computed. This will have no side effects other then that the ordering is now based on strings instead on integers, so the order will be different. This will however have no effect on existing configs as these will use the individual codes/keys and not the ordering to determine if there is a diff or not. Lastly (but I think also most importantly) there is a fix in this PR that makes diffing sets extremely more performand. Before a full diff required reading the complete Set for every single parameter/attribute you wanted to diff, while now it only gets that specific parameter. We have a use case where we have a Set that has 18 parameters and the set consist of about 600 items (don't ask :wink:). So when doing a diff it would take 100% CPU of all cores and stay that way for almost an hour before being able to complete the diff. Debugging this we learned that for retrieving every single parameter it made over 52.000 calls to `func (c *ResourceConfig) get(..)`. In this function a slice is created and used only for the duration of the call, so the time needed to create all needed slices and on the other hand the time the garbage collector needed to clean them up again caused the system to cripple itself. Next to that there are also some expensive reflect calls in this function which also claimed a fair amount of CPU time. After this fix the number of calls needed to get a single parameter dropped from 52.000+ to only 2! :smiley:
2015-11-18 11:24:04 +01:00
}, nil
}
// Build up the set from the list elements
2015-01-09 03:02:19 +01:00
for i, v := range raw.Value.([]interface{}) {
// Check if any of the keys in this item are computed
computed := r.hasComputedSubKeys(
fmt.Sprintf("%s.%d", strings.Join(address, "."), i), schema)
Change Set internals and make (extreme) performance improvements Changing the Set internals makes a lot of sense as it saves doing conversions in multiple places and gives a central place to alter the key when a item is computed. This will have no side effects other then that the ordering is now based on strings instead on integers, so the order will be different. This will however have no effect on existing configs as these will use the individual codes/keys and not the ordering to determine if there is a diff or not. Lastly (but I think also most importantly) there is a fix in this PR that makes diffing sets extremely more performand. Before a full diff required reading the complete Set for every single parameter/attribute you wanted to diff, while now it only gets that specific parameter. We have a use case where we have a Set that has 18 parameters and the set consist of about 600 items (don't ask :wink:). So when doing a diff it would take 100% CPU of all cores and stay that way for almost an hour before being able to complete the diff. Debugging this we learned that for retrieving every single parameter it made over 52.000 calls to `func (c *ResourceConfig) get(..)`. In this function a slice is created and used only for the duration of the call, so the time needed to create all needed slices and on the other hand the time the garbage collector needed to clean them up again caused the system to cripple itself. Next to that there are also some expensive reflect calls in this function which also claimed a fair amount of CPU time. After this fix the number of calls needed to get a single parameter dropped from 52.000+ to only 2! :smiley:
2015-11-18 11:24:04 +01:00
code := set.add(v, computed)
2015-01-09 03:02:19 +01:00
indexMap[code] = i
}
Change Set internals and make (extreme) performance improvements Changing the Set internals makes a lot of sense as it saves doing conversions in multiple places and gives a central place to alter the key when a item is computed. This will have no side effects other then that the ordering is now based on strings instead on integers, so the order will be different. This will however have no effect on existing configs as these will use the individual codes/keys and not the ordering to determine if there is a diff or not. Lastly (but I think also most importantly) there is a fix in this PR that makes diffing sets extremely more performand. Before a full diff required reading the complete Set for every single parameter/attribute you wanted to diff, while now it only gets that specific parameter. We have a use case where we have a Set that has 18 parameters and the set consist of about 600 items (don't ask :wink:). So when doing a diff it would take 100% CPU of all cores and stay that way for almost an hour before being able to complete the diff. Debugging this we learned that for retrieving every single parameter it made over 52.000 calls to `func (c *ResourceConfig) get(..)`. In this function a slice is created and used only for the duration of the call, so the time needed to create all needed slices and on the other hand the time the garbage collector needed to clean them up again caused the system to cripple itself. Next to that there are also some expensive reflect calls in this function which also claimed a fair amount of CPU time. After this fix the number of calls needed to get a single parameter dropped from 52.000+ to only 2! :smiley:
2015-11-18 11:24:04 +01:00
r.indexMaps[strings.Join(address, ".")] = indexMap
return FieldReadResult{
Value: set,
Exists: true,
Change Set internals and make (extreme) performance improvements Changing the Set internals makes a lot of sense as it saves doing conversions in multiple places and gives a central place to alter the key when a item is computed. This will have no side effects other then that the ordering is now based on strings instead on integers, so the order will be different. This will however have no effect on existing configs as these will use the individual codes/keys and not the ordering to determine if there is a diff or not. Lastly (but I think also most importantly) there is a fix in this PR that makes diffing sets extremely more performand. Before a full diff required reading the complete Set for every single parameter/attribute you wanted to diff, while now it only gets that specific parameter. We have a use case where we have a Set that has 18 parameters and the set consist of about 600 items (don't ask :wink:). So when doing a diff it would take 100% CPU of all cores and stay that way for almost an hour before being able to complete the diff. Debugging this we learned that for retrieving every single parameter it made over 52.000 calls to `func (c *ResourceConfig) get(..)`. In this function a slice is created and used only for the duration of the call, so the time needed to create all needed slices and on the other hand the time the garbage collector needed to clean them up again caused the system to cripple itself. Next to that there are also some expensive reflect calls in this function which also claimed a fair amount of CPU time. After this fix the number of calls needed to get a single parameter dropped from 52.000+ to only 2! :smiley:
2015-11-18 11:24:04 +01:00
}, nil
2015-01-09 03:02:19 +01:00
}
// hasComputedSubKeys walks through a schema and returns whether or not the
// given key contains any subkeys that are computed.
func (r *ConfigFieldReader) hasComputedSubKeys(key string, schema *Schema) bool {
prefix := key + "."
switch t := schema.Elem.(type) {
case *Resource:
for k, schema := range t.Schema {
if r.Config.IsComputed(prefix + k) {
return true
}
if r.hasComputedSubKeys(prefix+k, schema) {
return true
}
}
}
return false
}
// nestedConfigFieldReader is a funny little thing that just wraps a
// ConfigFieldReader to call readField when ReadField is called so that
// we don't recalculate the set rewrites in the address, which leads to
// an infinite loop.
type nestedConfigFieldReader struct {
Reader *ConfigFieldReader
}
func (r *nestedConfigFieldReader) ReadField(
address []string) (FieldReadResult, error) {
return r.Reader.readField(address, true)
}