No Extra Tableau Downtime: Metadata API for Proactive Knowledge Well being

In in the present day’s world, the reliability of knowledge options is all the pieces. Once we construct dashboards and reviews, one expects that the numbers mirrored there are appropriate and up-to-date. Based mostly on these numbers, insights are drawn and actions are taken. For any unexpected purpose, if the dashboards are damaged or if the numbers are incorrect — then it turns into a fire-fight to repair all the pieces. If the problems are usually not mounted in time, then it damages the belief positioned on the info staff and their options. 

However why would dashboards be damaged or have unsuitable numbers? If the dashboard was constructed appropriately the primary time, then 99% of the time the problem comes from the info that feeds the dashboards — from the info warehouse. Some attainable situations are:

• Few ETL pipelines failed, so the brand new information isn’t but in
• A desk is changed with one other new one 
• Some columns within the desk are dropped or renamed
• Schemas in information warehouse have modified
• And plenty of extra.

There may be nonetheless an opportunity that the problem is on the Tableau web site, however in my expertise, many of the occasions, it’s all the time because of some modifications in information warehouse. Although we all know the foundation trigger, it’s not all the time simple to start out engaged on a repair. There may be no central place the place you’ll be able to test which Tableau information sources depend on particular tables. When you’ve got the Tableau Knowledge Administration add-on, it may assist, however from what I do know, its exhausting to seek out dependencies of customized sql queries utilized in information sources.

However, the add-on is simply too costly and most firms don’t have it. The true ache begins when it’s a must to undergo all the info sources manually to start out fixing it. On high of it, you will have a string of customers in your head impatiently ready for a quick-fix. The repair itself won’t be troublesome, it could simply be a time-consuming one.

What if we may anticipate these points and establish impacted information sources earlier than anybody notices an issue? Wouldn’t that simply be nice? Properly, there’s a method now with the Tableau Metadata API. The Metadata API makes use of GraphQL, a question language for APIs that returns solely the info that you just’re inquisitive about. For more information on what’s attainable with GraphQL, do take a look at GraphQL.org.

On this weblog submit, I’ll present you ways to hook up with the Tableau Metadata API utilizing Python’s Tableau Server Shopper (TSC) library to proactively establish information sources utilizing particular tables, so to act quick earlier than any points come up. As soon as which Tableau information sources are affected by a selected desk, you may make some updates your self or alert the homeowners of these information sources in regards to the upcoming modifications to allow them to be ready for it.

Connecting to the Tableau Metadata API

Lets hook up with the Tableau Server utilizing TSC. We have to import in all of the libraries we would wish for the train!

### Import all required libraries
import tableauserverclient as t
import pandas as pd
import json
import ast
import re

With the intention to hook up with the Metadata API, you’ll have to first create a private entry token in your Tableau Account settings. Then replace the & with the token you simply created. Additionally replace along with your Tableau web site. If the connection is established efficiently, then “Linked” might be printed within the output window.

### Hook up with Tableau server utilizing private entry token
tableau_auth = t.PersonalAccessTokenAuth("", "", 
                                           site_id="")
server = t.Server("https://dub01.on-line.tableau.com/", use_server_version=True)

with server.auth.sign_in(tableau_auth):
        print("Linked")

Lets now get an inventory of all information sources which might be printed in your web site. There are a lot of attributes you’ll be able to fetch, however for the present use case, lets preserve it easy and solely get the id, identify and proprietor contact data for each information supply. This might be our grasp record to which we’ll add in all different data.

############### Get all of the record of knowledge sources in your Website

all_datasources_query = """ {
  publishedDatasources {
    identify
    id
    proprietor {
    identify
    e mail
    }
  }
}"""
with server.auth.sign_in(tableau_auth):
    end result = server.metadata.question(
        all_datasources_query
    )

Since I would like this weblog to be focussed on proactively establish which information sources are affected by a selected desk, I’ll not be going into the nuances of Metadata API. To higher perceive how the question works, you’ll be able to confer with a really detailed Tableau’s personal Metadata API documentation.

One factor to notice is that the Metadata API returns information in a JSON format. Relying on what you’re querying, you’ll find yourself with a number of nested json lists and it could actually get very tough to transform this right into a pandas dataframe. For the above metadata question, you’ll find yourself with a end result which would really like beneath (that is mock information simply to present you an concept of what the output seems like):

{
  "information": {
    "publishedDatasources": [
      {
        "name": "Sales Performance DataSource",
        "id": "f3b1a2c4-1234-5678-9abc-1234567890ab",
        "owner": {
          "name": "Alice Johnson",
          "email": "[email protected]"
        }
      },
      {
        "identify": "Buyer Orders DataSource",
        "id": "a4d2b3c5-2345-6789-abcd-2345678901bc",
        "proprietor": {
          "identify": "Bob Smith",
          "e mail": "[email protected]"
        }
      },
      {
        "identify": "Product Returns and Profitability",
        "id": "c5e3d4f6-3456-789a-bcde-3456789012cd",
        "proprietor": {
          "identify": "Alice Johnson",
          "e mail": "[email protected]"
        }
      },
      {
        "identify": "Buyer Segmentation Evaluation",
        "id": "d6f4e5a7-4567-89ab-cdef-4567890123de",
        "proprietor": {
          "identify": "Charlie Lee",
          "e mail": "[email protected]"
        }
      },
      {
        "identify": "Regional Gross sales Developments (Customized SQL)",
        "id": "e7a5f6b8-5678-9abc-def0-5678901234ef",
        "proprietor": {
          "identify": "Bob Smith",
          "e mail": "[email protected]"
        }
      }
    ]
  }
}

We have to convert this JSON response right into a dataframe in order that its simple to work with. Discover that we have to extract the identify and e mail of the proprietor from contained in the proprietor object. 

### We have to convert the response into dataframe for straightforward information manipulation

col_names = end result['data']['publishedDatasources'][0].keys()
master_df = pd.DataFrame(columns=col_names)

for i in end result['data']['publishedDatasources']:
    tmp_dt = {ok:v for ok,v in i.gadgets()}
    master_df = pd.concat([master_df, pd.DataFrame.from_dict(tmp_dt, orient='index').T])

# Extract the proprietor identify and e mail from the proprietor object
master_df['owner_name'] = master_df['owner'].apply(lambda x: x.get('identify') if isinstance(x, dict) else None)
master_df['owner_email'] = master_df['owner'].apply(lambda x: x.get('e mail') if isinstance(x, dict) else None)

master_df.reset_index(inplace=True)
master_df.drop(['index','owner'], axis=1, inplace=True)
print('There are ', master_df.form[0] , ' datasources in your web site')

That is how the construction of master_df would appear like:

As soon as we now have the principle record prepared, we will go forward and begin getting the names of the tables embedded within the information sources. In case you are an avid Tableau consumer, that there are two methods to choosing tables in a Tableau information supply — one is to straight select the tables and set up a relation between them and the opposite is to make use of a customized sql question with a number of tables to attain a brand new resultant desk. Due to this fact, we have to handle each the instances.

Processing of Customized SQL question tables

Under is the question to get the record of all customized SQLs used within the web site together with their information sources. Discover that I’ve filtered the record to get solely first 500 customized sql queries. In case there are extra in your org, you’ll have to use an offset to get the following set of customized sql queries. There may be additionally an possibility of utilizing cursor methodology in Pagination while you wish to fetch giant record of outcomes (refer right here). For the sake of simplicity, I simply use the offset methodology as I do know, as there are lower than 500 customized sql queries used on the positioning.

# Get the info sources and the desk names from all of the customized sql queries used in your Website

custom_table_query = """  {
  customSQLTablesConnection(first: 500){
    nodes {
        id
        identify
        downstreamDatasources {
        identify
        }
        question
    }
  }
}
"""

with server.auth.sign_in(tableau_auth):
    custom_table_query_result = server.metadata.question(
        custom_table_query
    )

Based mostly on our mock information, that is how our output would appear like:

{
  "information": {
    "customSQLTablesConnection": {
      "nodes": [
        {
          "id": "csql-1234",
          "name": "RegionalSales_CustomSQL",
          "downstreamDatasources": [
            {
              "name": "Regional Sales Trends (Custom SQL)"
            }
          ],
          "question": "SELECT r.region_name, SUM(s.sales_amount) AS total_sales FROM ecommerce.sales_data.Gross sales s JOIN ecommerce.sales_data.Areas r ON s.region_id = r.region_id GROUP BY r.region_name"
        },
        {
          "id": "csql-5678",
          "identify": "ProfitabilityAnalysis_CustomSQL",
          "downstreamDatasources": [
            {
              "name": "Product Returns and Profitability"
            }
          ],
          "question": "SELECT p.product_category, SUM(s.revenue) AS total_profit FROM ecommerce.sales_data.Gross sales s JOIN ecommerce.sales_data.Merchandise p ON s.product_id = p.product_id GROUP BY p.product_category"
        },
        {
          "id": "csql-9101",
          "identify": "CustomerSegmentation_CustomSQL",
          "downstreamDatasources": [
            {
              "name": "Customer Segmentation Analysis"
            }
          ],
          "question": "SELECT c.customer_id, c.location, COUNT(o.order_id) AS total_orders FROM ecommerce.sales_data.Clients c JOIN ecommerce.sales_data.Orders o ON c.customer_id = o.customer_id GROUP BY c.customer_id, c.location"
        },
        {
          "id": "csql-3141",
          "identify": "CustomerOrders_CustomSQL",
          "downstreamDatasources": [
            {
              "name": "Customer Orders DataSource"
            }
          ],
          "question": "SELECT o.order_id, o.customer_id, o.order_date, o.sales_amount FROM ecommerce.sales_data.Orders o WHERE o.order_status = 'Accomplished'"
        },
        {
          "id": "csql-3142",
          "identify": "CustomerProfiles_CustomSQL",
          "downstreamDatasources": [
            {
              "name": "Customer Orders DataSource"
            }
          ],
          "question": "SELECT c.customer_id, c.customer_name, c.phase, c.location FROM ecommerce.sales_data.Clients c WHERE c.active_flag = 1"
        },
        {
          "id": "csql-3143",
          "identify": "CustomerReturns_CustomSQL",
          "downstreamDatasources": [
            {
              "name": "Customer Orders DataSource"
            }
          ],
          "question": "SELECT r.return_id, r.order_id, r.return_reason FROM ecommerce.sales_data.Returns r"
        }
      ]
    }
  }
}

Identical to earlier than after we have been creating the grasp record of knowledge sources, right here additionally we now have nested json for the downstream information sources the place we would wish to extract solely the “identify” a part of it. Within the “question” column, the whole customized sql is dumped. If we use regex sample, we will simply seek for the names of the desk used within the question.

We all know that the desk names all the time come after FROM or a JOIN clause they usually usually observe the format ... The is optionally available and many of the occasions not used. There have been some queries I discovered which used this format and I ended up solely getting the database and schema names, and never the whole desk identify. As soon as we now have extracted the names of the info sources and the names of the tables, we have to merge the rows per information supply as there will be a number of customized sql queries utilized in a single information supply.

### Convert the customized sql response into dataframe
col_names = custom_table_query_result['data']['customSQLTablesConnection']['nodes'][0].keys()
cs_df = pd.DataFrame(columns=col_names)

for i in custom_table_query_result['data']['customSQLTablesConnection']['nodes']:
    tmp_dt = {ok:v for ok,v in i.gadgets()}

    cs_df = pd.concat([cs_df, pd.DataFrame.from_dict(tmp_dt, orient='index').T])

# Extract the info supply identify the place the customized sql question was used
cs_df['data_source'] = cs_df.downstreamDatasources.apply(lambda x: x[0]['name'] if x and 'identify' in x[0] else None)
cs_df.reset_index(inplace=True)
cs_df.drop(['index','downstreamDatasources'], axis=1,inplace=True)

### We have to extract the desk names from the sql question. We all know the desk identify comes after FROM or JOIN clause
# Word that the identify of desk will be of the format ..
# Relying on the format of how desk known as, you'll have to modify the regex expression

def extract_tables(sql):
    # Regex to match database.schema.desk or schema.desk, keep away from alias
    sample = r'(?:FROM|JOIN)s+((?:[w+]|w+).(?:[w+]|w+)(?:.(?:[w+]|w+))?)b'
    matches = re.findall(sample, sql, re.IGNORECASE)
    return record(set(matches))  # Distinctive desk names

cs_df['customSQLTables'] = cs_df['query'].apply(extract_tables)
cs_df = cs_df[['data_source','customSQLTables']]

# We have to merge datasources as there will be a number of customized sqls utilized in the identical information supply
cs_df = cs_df.groupby('data_source', as_index=False).agg({
    'customSQLTables': lambda x: record(set(merchandise for sublist in x for merchandise in sublist))  # Flatten & make distinctive
})

print('There are ', cs_df.form[0], 'datasources with customized sqls utilized in it')

After we carry out all of the above operations, that is how the construction of cs_df would appear like:

Processing of normal Tables in Knowledge Sources

Now we have to get the record of all of the common tables utilized in a datasource which aren’t part of customized SQL. There are two methods to go about it. Both use the publishedDatasources object and test for upstreamTables or use DatabaseTable and test for upstreamDatasources. I’ll go by the primary methodology as a result of I would like the outcomes at a knowledge supply stage (mainly, I would like some code able to reuse once I wish to test a selected information supply in additional element). Right here once more, for the sake of simplicity, as an alternative of going for pagination, I’m looping by every datasource to make sure I’ve all the pieces. We get the upstreamTables inside the sector object in order that needs to be cleaned out.

############### Get the info sources with the common desk names utilized in your web site

### Its finest to extract the tables data for each information supply after which merge the outcomes.
# Since we solely get the desk data nested beneath fields, in case there are a whole lot of fields 
# utilized in a single information supply, we'll hit the response limits and won't be able to retrieve all the info.

data_source_list = master_df.identify.tolist()

col_names = ['name', 'id', 'extractLastUpdateTime', 'fields']
ds_df = pd.DataFrame(columns=col_names)

with server.auth.sign_in(tableau_auth):
    for ds_name in data_source_list:
        question = """ {
            publishedDatasources (filter: { identify: """"+ ds_name + """" }) {
            identify
            id
            extractLastUpdateTime
            fields {
                identify
                upstreamTables {
                    identify
                }
            }
            }
        } """
        ds_name_result = server.metadata.question(
        question
        )
        for i in ds_name_result['data']['publishedDatasources']:
            tmp_dt = {ok:v for ok,v in i.gadgets() if ok != 'fields'}
            tmp_dt['fields'] = json.dumps(i['fields'])
        ds_df = pd.concat([ds_df, pd.DataFrame.from_dict(tmp_dt, orient='index').T])

ds_df.reset_index(inplace=True)

That is how the construction of ds_df would look:

We are able to must flatten out the fields object and extract the sector names in addition to the desk names. Because the desk names might be repeating a number of occasions, we must deduplicate to maintain solely the distinctive ones.

# Perform to extract the values of fields and upstream tables in json lists
def extract_values(json_list, key):
    values = []
    for merchandise in json_list:
        values.append(merchandise[key])
    return values

ds_df["fields"] = ds_df["fields"].apply(ast.literal_eval)
ds_df['field_names'] = ds_df.apply(lambda x: extract_values(x['fields'],'identify'), axis=1)
ds_df['upstreamTables'] = ds_df.apply(lambda x: extract_values(x['fields'],'upstreamTables'), axis=1)

# Perform to extract the distinctive desk names 
def extract_upstreamTable_values(table_list):
    values = set()a
    for inner_list in table_list:
        for merchandise in inner_list:
            if 'identify' in merchandise:
                values.add(merchandise['name'])
    return record(values)

ds_df['upstreamTables'] = ds_df.apply(lambda x: extract_upstreamTable_values(x['upstreamTables']), axis=1)
ds_df.drop(["index","fields"], axis=1, inplace=True)

As soon as we do the above operations, the ultimate construction of ds_df would look one thing like this:

We’ve got all of the items and now we simply must merge them collectively:

###### Be part of all the info collectively
master_data = pd.merge(master_df, ds_df, how="left", on=["name","id"])
master_data = pd.merge(master_data, cs_df, how="left", left_on="identify", right_on="data_source")

# Save the outcomes to analyse additional
master_data.to_excel("Tableau Knowledge Sources with Tables.xlsx", index=False)

That is our closing master_data:

Desk-level Impression Evaluation

Let’s say there have been some schema modifications on the “Gross sales” desk and also you wish to know which information sources might be impacted. Then you’ll be able to merely write a small operate which checks if a desk is current in both of the 2 columns — upstreamTables or customSQLTables like beneath.

def filter_rows_with_table(df, col1, col2, target_table):
    """
    Filters rows in df the place target_table is a part of any worth in both col1 or col2 (helps partial match).
    Returns full rows (all columns retained).
    """
    return df[
        df.apply(
            lambda row: 
                (isinstance(row[col1], record) and any(target_table in merchandise for merchandise in row[col1])) or
                (isinstance(row[col2], record) and any(target_table in merchandise for merchandise in row[col2])),
            axis=1
        )
    ]
# For example 
filter_rows_with_table(master_data, 'upstreamTables', 'customSQLTables', 'Gross sales')

Under is the output. You possibly can see that 3 information sources might be impacted by this transformation. It’s also possible to alert the info supply homeowners Alice and Bob upfront about this to allow them to begin engaged on a repair earlier than one thing breaks on the Tableau dashboards.

You possibly can take a look at the whole model of the code in my Github repository right here.

That is simply one of many potential use-cases of the Tableau Metadata API. It’s also possible to extract the sector names utilized in customized sql queries and add to the dataset to get a field-level impression evaluation. One may monitor the stale information sources with the extractLastUpdateTime to see if these have any points or have to be archived if they don’t seem to be used any extra. We are able to additionally use the dashboards object to fetch data at a dashboard stage.

Closing Ideas

When you’ve got come this far, kudos. This is only one use case of automating Tableau information administration. It’s time to mirror by yourself work and suppose which of these different duties you would automate to make your life simpler. I hope this mini-project served as an gratifying studying expertise to know the facility of Tableau Metadata API. In case you preferred studying this, you may additionally like one other certainly one of my weblog posts about Tableau, on a few of the challenges I confronted when coping with massive .