DBVoyeur: Your SQL under the microscope
By Oninit · Published 2 November 2026
Most database monitors ask the database what it is doing. They read system tables, poll counters or switch on tracing. That shows the database's own view of the work, and it costs the database something to produce it.
DBVoyeur takes the other side of the conversation. It watches the network traffic between the applications and the database, decodes the database's wire protocol, and records each SQL statement the way the application experienced it: when it was sent, how long it took to come back, and how much data came with the answer.
DBVoyeur supports all the major database engines, with Informix® as its primary database. Today, the free edition is available for Informix only.
What "passive capture" means
A capture agent runs on the machine that hosts the database. It listens to the traffic that machine sends and receives for the database's own port, using the standard packet-capture library. Because it runs on the server itself, it can also see connections that never touch the network, which the next section explains. It does not sit in the path of the traffic, it does not rewrite or delay a packet, and it does not log in to the database. If the agent stops, the database and its applications carry on as before.
Two things follow from that:
- No load inside the database. Nothing extra is executed in the engine, and no trace flag or audit setting needs to be turned on. The agent does use some CPU on the host, about 4% of one core at 40 statements a second.
- The application's point of view. The time between the request leaving the application and the reply arriving is what users wait for. It includes the network and the client as well as the engine, and a monitor that only sees the engine can miss where the time went.
From packets to statements
A packet stream is not a list of SQL statements. The agent has to follow each connection, put the pieces back in order, and understand the database's own protocol well enough to find the statement text, the parameters, the response and the end of the response. Each database engine speaks a different protocol, so each needs its own parser.
The parsed transactions are sent from the capture agent to a collector that stores them and serves a web dashboard. From there you can ask the questions that matter in practice: which statements are slowest, which ones are run most often, which applications and users send them, and how that changes over the day.
Capturing Informix shared-memory connections
Not every client reaches the database over the network. On Informix, a client running on the database server itself can connect through shared memory. That traffic exists only inside the machine, so a network capture alone would never see it.
Because DBVoyeur runs on the database host, it can capture these connections as well. For shared-memory clients it reads each statement inside the client process, at the moment the client hands it over to the database, using standard Linux kernel tracing. It is a read-only observation: the application is not modified or recompiled, and no proxy is put in front of it.
What to expect:
- Linux, on the database host. The capture uses kernel facilities available on current Linux distributions, and the traffic only exists inside that machine, so DBVoyeur has to run there.
- Switched on deliberately. Shared-memory capture is a separate capability. You enable it when you want it.
- Specific client programs. It observes shared-memory activity inside the client executables you configure, not every process on the server.
- Completed statements. A statement is recorded when its result returns. Very long statement text is truncated, and under sustained bursts far above normal rates a small number of events can be dropped.
- Java clients are unaffected. The Informix JDBC driver connects over the network, so Java applications cannot use shared memory and are captured as ordinary network traffic.
- Who ran it. For sessions that were already open, the user shown may be the connecting account and not the end user.
Architecture
DBVoyeur runs as three services. They are kept apart so that only the part that needs special rights has them.
- The capture agent. It runs on the database host, because it reads that host's own traffic and local shared-memory clients. It is the only privileged part: it needs raw network access on the host. It sends what it captures to the analysis server, connecting out to it.
- The analysis server and dashboard. An ordinary, unprivileged service. It receives the parsed transactions from the capture agent, owns the connection to the datastore, and serves the web dashboard. It can run on the database host or on another machine.
- The datastore. A repository built for very large volumes of per-transaction performance data, which you can search with SQL. Only the analysis server connects to it, and it is not published outside the deployment.
The dashboard is the only part that people connect to. It is served over HTTPS, on a self-signed certificate that is generated on first run and can be replaced with your own.
For a small installation, such as the free edition, all three services run together on one machine and are started with a single command. For a larger estate the same services are spread across machines, with TLS on the links between them. Configuration is supplied as environment variables, and prepared configuration files can be mounted to replace the generated ones.
Database client drivers are not shipped with DBVoyeur. They are needed only when you want it to collect statistics from the database itself in addition to capturing traffic, and in that case you place the driver for your engine in a directory that is mounted read-only into the service.
Practical notes
- Run it on the database server. It has to be on the host: mirror ports, SPAN and network taps are not supported.
- Encrypted connections. DBVoyeur does not terminate TLS or alter a byte between client and server. What it can report for an encrypted session depends on how the session is encrypted, so check which TLS version and cipher suite your listener negotiates before you assume what will be visible.
- Bound the logs. A monitoring tool must never become the outage. Whatever you run it with, set a size limit on container and application logs, so a fault in the tool cannot fill the disk of the machine it is watching.
- Decide who sees what. A tool that records every SQL statement also records the values inside them. Treat the dashboard and its datastore as sensitive, and give access only to the people who need it.
Common questions
Does DBVoyeur need a SPAN port or a network tap?
No. It runs on the database host and reads that host's own traffic. Mirror ports, SPAN and network taps are not supported.
Does it slow the database down?
It runs nothing extra inside the database engine. The capture agent does use some CPU on the host: about 4% of one core at 40 statements a second.
Does DBVoyeur work with TLS-encrypted connections?
It does not terminate TLS or change a byte between the client and the server. What it can report for an encrypted session depends on how the session is encrypted, so check which TLS version and cipher suite your listener negotiates before you assume what will be visible.
Is there a free edition?
Yes. Today it is available for Informix only, for one database, with 24 hours of retention.
Summary
Watching the wire gives you the application's view of every statement, without asking anything of the database. The price is a parser for each protocol and a clear understanding of what crosses the network. For performance work where "the database is fine but the application is slow" is the opening complaint, that view is often the one that was missing. To find out more, including the editions available, see the DBVoyeur product page, including the free Informix edition.