# Searching Workflows(Advanced visibility)

# Introduction

Cadence supports creating workflows with customized key-value pairs, updating the information within the workflow code, and then listing/searching workflows with a SQL-like query. For example, you can create workflows with keys city and age, then search all workflows with city = seattle and age > 22.

Also note that normal workflow properties like start time and workflow type can be queried as well. For example, the following query could be specified when listing workflows from the CLI or using the list APIs (Go (opens new window), Java (opens new window)):

WorkflowType = "main.Workflow" AND CloseStatus != "completed" AND (StartTime > 
   "2019-06-07T16:46:34-08:00" OR CloseTime > "2019-06-07T16:46:34-08:00") 
   ORDER BY StartTime DESC 

In other places, this is also called as advanced visibility. While basic visibility is referred to basic listing without being able to search.

# Memo vs Search Attributes

Cadence offers two methods for creating workflows with key-value pairs: memo and search attributes. Memo can only be provided on workflow start. Also, memo data are not indexed, and are therefore not searchable. Memo data are visible when listing workflows using the list APIs. Search attributes data are indexed so you can search workflows by querying on these attributes. However, search attributes require the use of Elasticsearch.

Memo and search attributes are available in the Go client in StartWorkflowOptions (opens new window).

type StartWorkflowOptions struct {
    // ...
    // Memo - Optional non-indexed info that will be shown in list workflow.
    Memo map[string]interface{}
    // SearchAttributes - Optional indexed info that can be used in query of List/Scan/Count workflow APIs (only
    // supported when Cadence server is using Elasticsearch). The key and value type must be registered on Cadence server side.
    // Use GetSearchAttributes API to get valid key and corresponding value type.
    SearchAttributes map[string]interface{}

In the Java client, the WorkflowOptions.Builder has similar methods for memo (opens new window) and search attributes (opens new window).

Some important distinctions between memo and search attributes:

  • Memo can support all data types because it is not indexed. Search attributes only support basic data types (including String(aka Text), Int, Float, Bool, Datetime) because it is indexed by Elasticsearch.
  • Memo does not restrict on key names. Search attributes require that keys are allowlisted before using them because Elasticsearch has a limit on indexed keys.
  • Memo doesn't require Cadence clusters to depend on Elasticsearch while search attributes only works with Elasticsearch.

# Search Attributes (Go Client Usage)

When using the Cadence Go client, provide key-value pairs as SearchAttributes in StartWorkflowOptions (opens new window).

SearchAttributes is map[string]interface{} where the keys need to be allowlisted so that Cadence knows the attribute key name and value type. The value provided in the map must be the same type as registered.

# Allow Listing Search Attributes

Start by querying the list of search attributes using the CLI

$ cadence --domain samples-domain cl get-search-attr
|         KEY         | VALUE TYPE |
| CloseStatus         | INT        |
| CloseTime           | INT        |
| CustomBoolField     | DOUBLE     |
| CustomDatetimeField | DATETIME   |
| CustomDomain        | KEYWORD    |
| CustomDoubleField   | BOOL       |
| CustomIntField      | INT        |
| CustomKeywordField  | KEYWORD    |
| CustomStringField   | STRING     |
| DomainID            | KEYWORD    |
| ExecutionTime       | INT        |
| HistoryLength       | INT        |
| RunID               | KEYWORD    |
| StartTime           | INT        |
| WorkflowID          | KEYWORD    |
| WorkflowType        | KEYWORD    |

Use the admin CLI to add a new search attribute:

cadence --domain samples-domain adm cl asa --search_attr_key NewKey --search_attr_type 1

The numbers for the attribute types map as follows:

  • 0 = String(Text)
  • 1 = Keyword
  • 2 = Int
  • 3 = Double
  • 4 = Bool
  • 5 = DateTime

# Keyword vs String(Text)

Note 1: String has been renamed to Text in ElasticSearch (opens new window). Cadence is also planning (opens new window) to rename it.

Note 2: Keyword and String(Text) are concepts taken from Elasticsearch. Each word in a String(Text) is considered a searchable keyword. For a UUID, that can be problematic as Elasticsearch will index each portion of the UUID separately. To have the whole string considered as a searchable keyword, use the Keyword type.

For example, key RunID with value "2dd29ab7-2dd8-4668-83e0-89cae261cfb1"

  • as a Keyword will only be matched by RunID = "2dd29ab7-2dd8-4668-83e0-89cae261cfb1" (or in the future with regular expressions (opens new window))
  • as a String(Text) will be matched by RunID = "2dd8", which may cause unwanted matches

Note: String(Text) type can not be used in Order By query.

There are some pre-allowlisted search attributes that are handy for testing:

  • CustomKeywordField
  • CustomIntField
  • CustomDoubleField
  • CustomBoolField
  • CustomDatetimeField
  • CustomStringField

Their types are indicated in their names.

# Value Types

Here are the Search Attribute value types and their correspondent Golang types:

  • Keyword = string
  • Int = int64
  • Double = float64
  • Bool = bool
  • Datetime = time.Time
  • String = string

# Limit

We recommend limiting the number of Elasticsearch indexes by enforcing limits on the following:

  • Number of keys: 100 per workflow
  • Size of value: 2kb per value
  • Total size of key and values: 40kb per workflow

Cadence reserves keys like DomainID, WorkflowID, and RunID. These can only be used in list queries. The values are not updatable.

# Upsert Search Attributes in Workflow

UpsertSearchAttributes (opens new window) is used to add or update search attributes from within the workflow code.

Go samples for search attributes can be found at github.com/uber-common/cadence-samples (opens new window).

UpsertSearchAttributes will merge attributes to the existing map in the workflow. Consider this example workflow code:

func MyWorkflow(ctx workflow.Context, input string) error {
    attr1 := map[string]interface{}{
        "CustomIntField": 1,
        "CustomBoolField": true,
    workflow.UpsertSearchAttributes(ctx, attr1)
    attr2 := map[string]interface{}{
        "CustomIntField": 2,
        "CustomKeywordField": "seattle",
    workflow.UpsertSearchAttributes(ctx, attr2)

After the second call to UpsertSearchAttributes, the map will contain:

    "CustomIntField": 2,
    "CustomBoolField": true,
    "CustomKeywordField": "seattle",

There is no support for removing a field. To achieve a similar effect, set the field to a sentinel value. For example, to remove “CustomKeywordField”, update it to “impossibleVal”. Then searching CustomKeywordField != ‘impossibleVal’ will match workflows with CustomKeywordField not equal to "impossibleVal", which includes workflows without the CustomKeywordField set.

Use workflow.GetInfo to get current search attributes.

# ContinueAsNew and Cron

When performing a ContinueAsNew or using Cron, search attributes (and memo) will be carried over to the new run by default.

# Query Capabilities

Query workflows by using a SQL-like where clause when listing workflows from the CLI or using the list APIs (Go (opens new window), Java (opens new window)).

Note that you will only see workflows from one domain when querying.

# Supported Operators

  • AND, OR, ()
  • =, !=, >, >=, <, <=
  • IN

# Default Attributes

More and more default attributes are added in newer versions. Please get the by using the CLI get-search-attr command or the GetSearchAttributes API. Some names and types are as follows:

CloseStatus INT
CloseTime INT
CustomBoolField DOUBLE
CustomDatetimeField DATETIME
CustomDomain KEYWORD
CustomDoubleField BOOL
CustomIntField INT
CustomKeywordField KEYWORD
CustomStringField STRING
ExecutionTime INT
HistoryLength INT
StartTime INT
WorkflowType KEYWORD
Tasklist KEYWORD

There are some special considerations for these attributes:

  • CloseStatus, CloseTime, DomainID, ExecutionTime, HistoryLength, RunID, StartTime, WorkflowID, WorkflowType are reserved by Cadence and are read-only
  • Starting from v0.18.0 (opens new window), Cadence automatically maps(case insensitive) string to CloseStatus so that you don't need to use integer in the query, to make it easier to use.
    • 0 = "completed"
    • 1 = "failed"
    • 2 = "canceled"
    • 3 = "terminated"
    • 4 = "continued_as_new"
    • 5 = "timed_out"
  • StartTime, CloseTime and ExecutionTime are stored as INT, but support queries using both EpochTime in nanoseconds, and string in RFC3339 format (ex. "2006-01-02T15:04:05+07:00")
  • CloseTime, CloseStatus, HistoryLength are only present in closed workflow
  • ExecutionTime is for Retry/Cron user to query a workflow that will run in the future
  • To list only open workflows, add CloseTime = missing to the end of the query.

If you use retry or the cron feature to query workflows that will start execution in a certain time range, you can add predicates on ExecutionTime. For example: ExecutionTime > 2019-01-01T10:00:00-07:00. Note that if predicates on ExecutionTime are included, only cron or a workflow that needs to retry will be returned.

# General Notes About Queries

  • Pagesize default is 1000, and cannot be larger than 10k
  • Range query on Cadence timestamp (StartTime, CloseTime, ExecutionTime) cannot be larger than 9223372036854775807 (maxInt64 - 1001)
  • Query by time range will have 1ms resolution
  • Query column names are case sensitive
  • ListWorkflow may take longer when retrieving a large number of workflows (10M+)
  • To retrieve a large number of workflows without caring about order, use the ScanWorkflow API
  • To efficiently count the number of workflows, use the CountWorkflow API

# Tools Support


Support for search attributes is available as of version 0.6.0 of the Cadence server. You can also use the CLI from the latest CLI Docker image (opens new window) (supported on 0.6.4 or later).

# Start Workflow with Search Attributes

cadence --do samples-domain workflow start --tl helloWorldGroup --wt main.Workflow --et 60 --dt 10 -i '"vancexu"' -search_attr_key 'CustomIntField | CustomKeywordField | CustomStringField |  CustomBoolField | CustomDatetimeField' -search_attr_value '5 | keyword1 | vancexu test | true | 2019-06-07T16:16:36-08:00'

# Search Workflows with List API/Command

cadence --do samples-domain wf list -q '(CustomKeywordField = "keyword1" and CustomIntField >= 5) or CustomKeywordField = "keyword2"' -psa
cadence --do samples-domain wf list -q 'CustomKeywordField in ("keyword2", "keyword1") and CustomIntField >= 5 and CloseTime between "2018-06-07T16:16:36-08:00" and "2019-06-07T16:46:34-08:00" order by CustomDatetimeField desc' -psa

To list only open workflows, add CloseTime = missing to the end of the query.

Note that queries can support more than one type of filter:

cadence --do samples-domain wf list -q 'WorkflowType = "main.Workflow" and (WorkflowID = "1645a588-4772-4dab-b276-5f9db108b3a8" or RunID = "be66519b-5f09-40cd-b2e8-20e4106244dc")'
cadence --do samples-domain wf list -q 'WorkflowType = "main.Workflow" StartTime > "2019-06-07T16:46:34-08:00" and CloseTime = missing'

All above command can be done with ListWorkflowExecutions API.

# Count Workflows with Count API/Command

cadence --do samples-domain wf count -q '(CustomKeywordField = "keyword1" and CustomIntField >= 5) or CustomKeywordField = "keyword2"'
cadence --do samples-domain wf count -q 'CloseStatus="failed"'
cadence --do samples-domain wf count -q 'CloseStatus!="completed"'

All above command can be done with CountWorkflowExecutions API.

# Web UI Support

Queries are supported in Cadence Web (opens new window) as of release 3.4.0. Use the "Basic/Advanced" button to switch to "Advanced" mode and type the query in the search box.

# TLS Support for connecting to Elasticsearch

If your elasticsearch deployment requires TLS to connect to it, you can add the following to your config template. The TLS config is optional and when not provided it defaults to tls.enabled to false

    scheme: "https"
    host: ""
    visibility: cadence-visibility-dev
    enabled: true
    caFile: /secrets/cadence/elasticsearch_cert.pem
    enableHostVerification: true
    serverName: myServerName
    certFile: /secrets/cadence/certfile.crt
    keyFile: /secrets/cadence/keyfile.key
    sslmode: false

# Running Locally

  1. Increase Docker memory to higher than 6GB. Navigate to Docker -> Preferences -> Advanced -> Memory
  2. Get the Cadence Docker compose file. Run curl -O https://raw.githubusercontent.com/uber/cadence/master/docker/docker-compose-es.yml
  3. Start Cadence Docker (which contains Apache Kafka, Apache Zookeeper, and Elasticsearch) using docker-compose -f docker-compose-es.yml up
  4. From the Docker output log, make sure Elasticsearch and Cadence started correctly. If you encounter an insufficient disk space error, try docker system prune -a --volumes
  5. Register a local domain and start using it. cadence --do samples-domain d re
  6. Add the key to ElasticSearch And also allowlist search attributes. cadence --do domain adm cl asa --search_attr_key NewKey --search_attr_type 1

# Running in Production

To enable this feature in a Cadence cluster:

To add new search attributes:

  1. Add the key to ElasticSearch cadence --do domain adm cl asa --search_attr_key NewKey --search_attr_type 1
  2. Update the dynamic configuration (opens new window) to allowlist the new attribute

Note: starting a workflow with search attributes but without advanced visibility feature will succeed as normal, but will not be searchable and will not be shown in list results.