Practical writing on data integration from the team building production-grade adapters: ETL vs ELT, webhooks vs polling, data mapping rules, and what point-to-point scripts really cost. No filler, plenty of field names.
Cost
Informatica publishes no price per IPU on its own site, but it lists IDMC on AWS Marketplace at 120 IPUs a month for $131,760 on a 12-month contract, about $91.50 per IPU-month at list. The conversion from PowerCenter is metered against the same pool. Here is every line of a PowerCenter to IDMC budget, which ones Informatica prices, and which ones it does not.
9 min read
Migration
Most MySQL types have an obvious Snowflake target and a connector will pick it without complaint. The dangerous ones convert, load, reconcile perfectly on row counts, and change what your data means. MySQL BOOL is TINYINT and arrives as a number, so a column holding 2 is excluded by a = TRUE filter in MySQL and included by the same filter in Snowflake. Here is every mapping, plus the one line query that proves each risky one is safe, run against MySQL before the load.
12 min read
Migration
Snowflake publishes two Oracle to Snowflake type mappings and they disagree about the most common numeric column in an Oracle schema. A NUMBER declared with no precision becomes NUMBER(38,19) under one and NUMBER(38,18) under the other, and that one digit decides whether a twenty digit value loads or overflows. Here is every mapping, every disagreement adjudicated against Oracle's own reference, and the query that proves each risky conversion is safe, run against Oracle before the load.
12 min read
Migration
Snowflake publishes two separate PostgreSQL type mappings and they disagree on timestamps, arrays and currency. One of them sends a money column to FLOAT, which is the exact thing the PostgreSQL manual tells you not to do. Here is every mapping, plus the query that proves each risky one is safe, run against Postgres before the load.
12 min read
Migration
Most SQL Server types have an obvious Snowflake target, and a converter will pick it without complaint. Roughly a third of those conversions then produce valid DDL, load without an error, reconcile on row counts, and change what your data means. Here is every mapping, plus the one line query that proves each risky one is safe, run against SQL Server before the load.
12 min read
Migration
Most T-SQL converts to PostgreSQL mechanically. TOP becomes LIMIT, IIF becomes a CASE expression, and a converter handles the bulk of it. Then comes the part that costs teams a fortnight of debugging after go-live: six translations that every conversion guide publishes, which compile, run, return a value, and return a different value than the statement they replaced.
11 min read
Migration
SnowConvert converts Netezza tables and views, and stops there. The NZPLSQL procedures that run the nightly jobs are a rewrite into Snowflake Scripting. Most constructs map one to one. Six do not, from REFTABLE result sets to division that keeps fewer decimal places, and each compiles and returns a different answer.
10 min read
Migration
Converters handle most of an Informix code base: DEFINE, LET, FOREACH and ON EXCEPTION all have PL/pgSQL equivalents. Six SPL behaviors do not survive the trip, from ON EXCEPTION WITH RESUME to DATETIME variables declared as DATE, and each one compiles in PostgreSQL and returns a different answer.
10 min read
Migration
Converters handle most of an SAP ASE code base: @@rowcount, raiserror, *= joins and getdate() all have PL/pgSQL equivalents. Six ASE behaviors do not survive the trip, from blank strings stored as a space to set rowcount on DELETE, and each one compiles in PostgreSQL and returns a different answer.
9 min read
Migration
Most Db2 types cross into PostgreSQL unchanged. DECFLOAT, long timestamps, FOR BIT DATA, oversized LOBs and identity columns do not. A full type mapping, the defaults to reject, and the SYSCAT query to run on Db2 that proves each risky mapping safe before a single row loads.
10 min read
Migration
Converting Oracle DDL to MySQL looks mechanical until you meet NUMBER, DATE and the empty string. A full type mapping for every common Oracle column, the four decisions no converter can make for you, and the query to run on Oracle that proves each risky mapping safe before a single row loads.
10 min read
Migration
Most SQL Server to MySQL type conversions are obvious. About ten are not, and they deploy cleanly while changing what the data means: a rowversion mapped to a date, MONEY to a whole number, datetime2 rounded across midnight. The full mapping table, plus the query to run on SQL Server that proves each risky mapping safe before the load.
10 min read
Migration
Most MySQL to PostgreSQL type conversions are obvious and a converter gets them right. About a dozen are not, and those produce a schema that deploys without a single error and quietly means something different. The full mapping table, plus the part nobody publishes: the one line query that proves each risky mapping is safe, run against MySQL before the load rather than PostgreSQL after it.
10 min read
Migration
AWS publishes an automation rating for every Oracle feature area in its own migration playbook, and two of them are rated no automation at all: MERGE statements and database links. Ora2Pg calls its own PL/SQL conversion basic. Here is what actually converts, what you rewrite by hand, and the shorter list that worries experienced teams most: the constructs that convert at the highest rating and quietly return different results.
10 min read
Cost
The tool is almost never the expensive part of a data migration. Cloud services bill per replication hour, ELT vendors bill per row, consultancies bill per day, and all of it is dwarfed by the time people spend mapping fields and proving the result is correct. Here is how each pricing model behaves when a project runs long, and how to build an estimate that survives contact with the data.
9 min read
Integration
A NetSuite account on the Standard service tier gets five concurrent web services requests in total, shared by every integration connected to it. That single number breaks more NetSuite projects than any rate limit, and the error it produces is HTTP 400 rather than the 429 every retry library watches for. The real limits by tier, the six failures teams experience as one error, and the fixes.
11 min read
Integration
Stripe allows 100 requests a second in live mode and 25 in a sandbox. Those are the numbers everyone quotes. The reason Stripe integrations break is that six genuinely different problems all return HTTP 429, and their fixes contradict each other. The current limits, how to tell the six apart in one header, and seven changes that fix them.
12 min read
Integration
QuickBooks throttles at 500 requests a minute per company file and allows ten concurrent connections to it. Almost every 429 we get asked about is the second limit, not the first, which is why adding workers makes a slow QuickBooks sync slower. The current numbers, the five failures that all look like one problem, and the six changes that fix them.
11 min read
Integration
Shopify does not count your requests, it prices them. The GraphQL Admin API meters calculated query cost in points per second and your plan sets the rate, which is why the same connector behaves like a different product on Standard and on Plus. The current numbers, and the six changes that fix a throttled sync.
11 min read
Integration
Almost every stalled HubSpot integration is stalled for one of two reasons, and neither is the daily call limit everybody worries about. The current numbers by tier, what a 429 actually costs you, and the six changes that fix a rate-limited sync.
10 min read
Integration
Enterprise, Unlimited and Performance orgs get five of these at no extra cost, and most of them sit unclaimed while an integration runs as somebody's admin account. What the license permits, what it costs, where it gets in your way, and the setup order that trips up almost every first attempt.
10 min read
Databases
Zero-ETL is described in two wrong ways: as marketing for a faster connector, or as a universal replacement for ETL. It is neither. Here is exactly what AWS replicates for you, what you give up in exchange, and the failures that show up quietly in week one.
11 min read
Databases
The commands are three lines long and teams still end up throttled by a quota nobody read, money columns that lost their cents, and a pipeline that stopped a month ago. Every realistic path into BigQuery, with the numbers from Google's own documentation.
11 min read
Databases
The commands are short and almost every team still ends up with slow loads, money columns that lost their cents, and a pipeline that stopped three weeks ago. Every realistic path into Snowflake, with the numbers from Snowflake's own documentation.
11 min read
Integrations
Two completely different things get called connecting Snowflake to Salesforce, and picking the wrong one costs a month. Here is the difference between a write that lands on records and zero copy federation that does not, plus the API limits and type mismatches that break the first sync.
10 min read
Databases
PostgreSQL ships real logical replication, and it is the right answer whenever both ends are Postgres. It also silently refuses to replicate your schema, your sequences and your large objects. Here is the setup, the documented restrictions, and the slot that fills your disk.
10 min read
Databases
PostgreSQL gives you real log-based CDC through logical decoding, and it gives you a replication slot that will retain write-ahead log files until your disk is full. Here is how the pieces fit, what to set on RDS, and when a watermark query is the better answer.
9 min read
Integration
Stripe charges into QuickBooks with the processor fee on its own account. A Shopify order that decrements NetSuite inventory before the warehouse oversells it. Twelve integrations teams actually run, and the three disagreements that make each one hard.
9 min read
RevOps
The strongest HubSpot to Snowflake route is HubSpot's own Data Share, and it is available on one subscription tier only. Below that tier, Snowflake's connector lands a JSON payload with a view of fields it considers common. Here is every option scored on the three questions that decide the purchase: custom objects, deleted records and property history.
9 min read
Data engineering
Every MongoDB to Snowflake tool reads the same change stream, so the pipeline is rarely what you are buying. What separates them is what happens to a nested document on arrival. Snowflake's own connector lands it whole in one column and leaves the modeling to you. Here is which tools do the flattening, and what each does when a field changes shape.
10 min read
Buying guides
Estuary, Debezium, DMS, Airbyte and Fivetran all sync MongoDB into MySQL, and two of them need a server setting MySQL ships disabled. The bigger risk is MySQL's own upsert, which matches on any unique index, not only the key.
8 min read
Buying guides
AWS DMS cannot read DynamoDB, so most AWS teams end up on Glue. The export connector spares production read capacity, a run costs a few dollars, and Glue's own transform leaves every number as a string. Here is the job, the bill, and the alternatives.
8 min read
Buying guides
Fivetran reads DynamoDB with a full scan, then DynamoDB Streams, and by default lands every item as one JSON column. Here is what that means in Snowflake, what the sync bills under 2026 pricing, what happens after a 24-hour outage, and which alternatives beat it for which job.
8 min read
Databases
MongoDB lands in BigQuery as JSON, with dates, longs and decimals wrapped in $date, $numberLong and $numberDecimal. Google changed the Datastream default on 16 September 2026, so a date now sits at a different path in new streams. Here is the SQL that reads both shapes and the query that shows which one each table holds.
9 min read
Databases
The popular MongoDB to PostgreSQL tools guess your schema from a sample and skip what the sample missed. The fix is nine short mongosh queries run against MongoDB before the load. Here is the mapping, the checks, and what each one catches.
10 min read
Buying guides
Google's HubSpot connector for BigQuery is free while it is in Preview, and it reloads all of your HubSpot data on every run. That makes it free today and the one whose bill grows with portal size once Google starts charging slot-hours. Here is what every route costs on the same portal, including the HubSpot API allowance nobody invoices.
9 min read
Buying guides
Google's own Salesforce connector for BigQuery is paid and publicly priced: $0.06 per slot-hour, with a planning ceiling of about $1.20 per hour of run time. That makes the schedule, not the row count, the thing that sets your bill. Here is what each route costs on the same Salesforce org.
9 min read
Data engineering
There is no single price for moving Salesforce into Snowflake, because there is no single product. Snowflake says its own connector is built for teams who do not own Salesforce Data Cloud, so a licensing fact picks your architecture before any feature comparison. Here is what each route actually bills you for.
9 min read
Buying guides
Qlik retired Talend Open Studio and moved Talend onto capacity-based cloud tiers with no public list price. Here is how the capacity model works, what contracts actually run, and what pushes the number up.
9 min read
Buying guides
Matillion sells the Data Productivity Cloud on a credit-based consumption model with no public list price. Here is how credits work, what the Developer, Teams, and Scale editions include, what a deployment runs, and what pushes the bill up.
9 min read
Buying guides
SnapLogic sells three packages with unlimited pipelines and no posted prices. Here is what the pricing page states, what its $125,000 AWS Marketplace listing works out to, and the questions that make a quote comparable.
8 min read
Buying guides
Informatica sells IDMC through sales on an IPU consumption model with no public list price. Here is how the meter works, the one price Informatica publishes, what PowerCenter adds, and what inflates the bill.
9 min read
Buying guides
Jitterbit publishes no prices, but it does publish how its plans are sized. Here are the connection, agent, and environment limits for each plan, where App Builder fits, and what to ask for in a quote.
8 min read
Buying guides
Tray.ai publishes plans but no prices, and bills usage in Tasks. Here is what each plan includes, what its $150,000 AWS Marketplace listing works out to, why the Task counting rule matters, and what to ask sales.
8 min read
Buying guides
Fivetran charges each distinct primary key once a month. Airbyte charges the volume moved, and a full refresh recharges every row. At high volume that one difference decides the winner.
9 min read
Buying guides
Airbyte Cloud Standard is $20 a month with 5 credits, and every extra credit is $5. Here is what each plan includes, what a million rows costs, and what happens when the credits run out.
11 min read
Buying guides
n8n cloud bills by workflow execution and the community edition is free to download. Here is what each tier actually covers, what pushes the meter up, and what self-hosting really costs to run.
8 min read
Buying guides
Celigo prices by endpoints and flows across Standard, Professional, and Enterprise, and publishes no list price. Here is what each edition includes and what to ask before you sign.
9 min read
Buying guides
Boomi publishes one price, Pay-As-You-Go at $99 a month plus $0.05 per message, and quotes every edition above it. Here is how both meters work, what drives the bill, and where a flat price wins.
8 min read
Buying guides
Fivetran charges by Monthly Active Rows and prints no rate per million on its pricing page. Here is what it does publish: the Free allowance, the $5 base charge, deletes counting since January 2026, annual discounts, and how to get a comparable quote.
9 min read
Buying guides
Self-serve platforms run tens to hundreds of dollars a month, enterprise suites run six figures a year, and in-house looks free until you price the maintenance. The arithmetic, in full.
9 min read
Guides
The build is the cheap part. Auth, pagination, retries, backfills, and every upstream API version bump are the bill you pay forever. A framework for deciding which side you are on.
9 min read
Data engineering
ETL transforms data before it lands, ELT transforms it after. The right choice depends on where your queries run and who maintains the pipeline.
7 min read
Guides
iPaaS is a hosted platform that moves data between your apps so you do not write and babysit point-to-point scripts. Here is how it works and when it pays off.
11 min read
Guides
An API adapter sits between two systems and translates one data shape into another. It is the piece of glue code every team rewrites, and the piece we productized.
6 min read
Data engineering
Most sync failures are mapping failures. Nine rules, from picking a stable match key to versioning every change, that keep records landing where they should.
8 min read
Data engineering
Webhooks are fast but drop events; polling is reliable but never instant. Production-grade integrations use both. Here is the decision framework.
9 min read
For decision makers
Every pair of connected systems adds another script someone has to own. At ten systems you are maintaining dozens of fragile links. The math gets ugly fast.
7 min read
Connect your first pair and watch records land where they should. Flat price from $49 a month.
The live demo needs no card, and Starter is $49 a month.
Confirm your email and start connecting sources.