Known limits
Every limit of what Straza protects in one table, with what it means for you, what reduces it today and what the roadmap plans, and the scale ceilings of one deployment.
On this page
Straza leaves some risks with you, and this page names each one as it stands in 2.2.0. A security reviewer or an auditor reads the table as the list of residual risks to accept, reduce or track. Each row gives the limit and the mechanism behind it, what it means for your deployment, what reduces it today, and what is planned. The link in each row leads to the page that explains the mechanism in full. The Planned column quotes only what the roadmap states, and a row the roadmap does not name says so.
| Limit | What it means for you | What reduces it today | Planned |
|---|---|---|---|
| A user-mode hook is advisory. The hook runs as the user, and an agent that can run anything on its machine can run around it. | A local tool rule on such a machine stops a mistake or a naive prompt injection. It does not hold against an agent that sets out to bypass it. | A managed install in root-owned paths, with the managed level required for sensitive roles through governance.minAttestation or a policy require block. The sandbox image for processes without a hook. |
Not on the roadmap. |
| An MCP server added outside the gateway. A server a person registers in their own harness never reaches the gateway, and only the advisory hook sees its calls. | Its credential sits on that machine, and governing it is that person’s responsibility. | Register every MCP server that holds a credential in Straza, so the gateway decides its calls and keeps the credential. On Claude Code, a managed install writes an exclusive managed MCP file, so the harness loads only the servers named there. | Not on the roadmap. |
The attestation level and the key’s custody come from the client. The server compares the hashes a check-in reports with its registry, and governance.minCustody checks the custody the client reports when it enrolls a person’s machine key. The server cannot tell which program sent either report. |
On Linux, macOS and Windows, straza service measures its own binary and the managed files under its own OS account, so a process of the person cannot make it report other hashes. A program of the person’s own, given a fresh sign-in of the person, can still enroll a key it keeps itself and report the registered hashes and the account custody. |
Treat managed and the reported custody as guards against a changed or wrong install. Never make either the only control in front of an action that must not happen. |
A device-bound client certificate to complement attestation. |
| A sandbox session cannot prove its image. A check-in reports whether the install is managed, its platform and its hashes, and nothing about the image it ran in. | That a session ran inside the sandbox image is your deployment’s claim. | Keep high-blast-radius permissions on roles that require managed, and put MCP tools behind the gateway. |
Attestation of the image a session ran in. |
| The sandbox image leaves two paths open. Every binary on its allowlist runs without a decision, and the shipped compose file leaves the container’s network routable. | A per-argument rule holds at the boundary only for a tool that is absent from the image. | Keep the allowlist as shipped, put tools whose arguments matter behind the gateway, and set internal: true on the sandbox network with only strazad attached. |
Not on the roadmap. |
| Interpreter indirection. An agent can write a script and then run it, and the hook sees only the interpreter call. | A rule that matches a command’s text does not see what the script does. | A rule can deny the tagged interpreter call or send it to the heuristic classifier. The audit sentinel, once governance.sentinel.enabled is true, records a written file that is then run, after the fact. |
Not on the roadmap. |
| What Straza does not try to do. Root on an endpoint owns every process on it, the hook included. Computation inside a process is ungoverned, and model traffic is never proxied. | No enforcement on a machine holds against its own administrator. Straza sees no prompt or completion unless a policy turns on conversation recording at the hook. | Keep credentials behind the gateway, where they never reach the machine, and govern the actions that leave the process. | None, by design. |
The service keeps each key in a plain file. straza service keeps each enrolled user’s Ed25519 key as a PKCS #8 file with mode 0600 in its state directory, which belongs to its own OS account, and never encrypts it or keeps it in a TPM or a keychain. A service started by hand with STRAZA_HOME runs as the person and keeps the key in the person’s own account. |
Root, an administrator on Windows, and anyone with the disk or a backup of it can read every key. A copy starts sessions from another machine until the original machine’s next check-in, when strazad sees the counter repeat and revokes the device, or until the sign-in deadline. Every process of a person whose service runs as them can read that key, an agent included. | Disk encryption, and the system service for every person. The enterprise default of governance.minCustody, account, refuses a person’s key reported as kept in their own account. Revoke the device when a machine or its backup is lost. |
Not on the roadmap. |
An OS account a person enrolled cannot pass to an agent. straza service refuses an agent’s key and a headless enrollment for an OS account that holds a person’s enrollment, so an agent that runs as the person cannot replace it. No command hands the account from the person to an agent. |
An OS account that a person once enrolled runs no headless agent until an administrator clears it. | Run each agent under an OS account of its own. To reuse an account that a person enrolled, an administrator resets that account’s state in straza service, as the section Hand an enrolled OS account to an agent in Headless agents shows. |
Not on the roadmap. |
An admin API token is a bearer secret. strazactl keeps no login and runs only on a token in STRAZA_API_TOKEN, so no admin credential sits on a machine unless someone puts a token there. Nothing binds a token to a machine or a person. |
Whoever holds a copy acts with its scope from any machine until it expires or is revoked. A token left in a shell profile, a file or a terminal an agent shares is the agent’s too. Disabling or locking the person who minted a token leaves the token working. | Never put a token where an agent runs as you. Give a token for a terminal narrow scopes and the shortest lifetime the console offers, 30 days, and revoke it under Settings, API tokens when the work is done. When a person leaves, revoke the tokens that list names as minted by them. Keep an automation token in your secret store. | Not on the roadmap. |
A copied bearer session token can be refreshed. A session token lives 300 seconds. A token that straza service holds is bound to its key, so every route needs a fresh proof from that key with it. The token of an AI agent that checks in with its key or client secret without the service is bound to nothing, and its holder can refresh it like the original while the session and the user stay active. |
A copied bearer token keeps working until you revoke the session, until it goes 600 seconds without a refresh, or until governance.sessionMaxLifetime ends the session, 12 hours by default. A copied token of the service works nowhere else. |
strazactl sessions revoke, disabling the user, or a shorter session lifetime. Give each agent a key the service holds, and rotate an identity provider client secret that sat in an agent’s environment. |
Sessions bound to the key of straza service cover Linux, macOS and Windows today. |
A revocation reaches a session late while its push stream is down. straza service holds a push stream for each live session and marks the session revoked when a revocation arrives on it. While the stream is down, the hooks keep deciding on the session the service holds until a call renews it, once 120 seconds or less of its 300-second token remain. |
A revoked session that keeps working can keep running local tools for up to about 180 seconds after its token was minted. A standalone service that cannot reach strazad adds its offline grace of 900 seconds. The gateway refuses a revoked token at once. | Let /v1/push through every proxy between the machines and strazad with no buffering, watch the killswitch line of straza doctor, and keep governance.offlineGraceTTL at 0, the enterprise default. |
Not on the roadmap. |
Snapshot keys are trusted on first use. straza enroll pins the snapshot signing keys the server publishes, over whatever connection it runs on. |
An enrollment over a connection someone else controls can pin their key, and the machine then accepts their policy. | Enroll over TLS or on a network you trust. | Not on the roadmap. |
Sign-in routes carry one rate limit. server.loginPerIPRPS, 2 requests a second per client address, covers the password submits of both built-in sign-in pages and, in enterprise, the agent token endpoint. The start of a device code sign-in has the same rate from a budget of its own, and strazad holds at most 4096 device code sign-ins at once. /v1/enroll and /v1/checkin carry none, and no account locks after failed attempts. |
An online guesser is slowed by bcrypt and that limit alone. Behind a proxy every client shares the proxy’s address and its budget. Four IPv4 addresses, or four IPv6 /64 networks, can fill the 4096 places, since an IPv6 client counts by its /64. While such a flood lasts strazad refuses new device code sign-ins, and the browser sign-in keeps working and the table stays at 4096 sign-ins. | Rate-limit the sign-in routes at your ingress, and alert on the login failures that the audit chain records. | Not on the roadmap. |
| Standalone signs in with passwords. The built-in issuer checks a password hashed with bcrypt and asks for no second factor. | A standalone account is as strong as its password. | The enterprise profile, where your identity provider owns the login and its second factor. | Passkeys for the standalone issuer. |
First-boot passwords land in the server log. The first boot writes the bootstrap admin password, in standalone, and the break-glass password, in both profiles, as the password field of a warning line. |
Every copy of that log, in a log shipper or a kept container log, holds full administration until the password changes. | Change both passwords after the first boot with strazactl users set-password, and keep the first boot’s log out of shared log stores. |
Not on the roadmap. |
| An agent can enroll an approval device as its person. An agent can use a sign-in of its person that it can read, such as the person’s unexpired ID token, and the server stores what an enrolling device reports without checking it. | A device an agent enrolled decides every request its person may decide, the agent’s own included. A signed decision proves which device decided, never that a person acted. | Keep every credential of a person away from their agents. Review strazactl approvers list and revoke a device nobody recognizes. Keep approval.unsignedOwnDecisions at false. |
A confirmation on a separate device before a new approval device can decide. |
| Whoever may set a person’s password can enroll an approval device as them. No admin route or console screen enrolls a device for someone else, and the self route enrolls one for whoever is signed in as the person. On the standalone profile strazad is the identity provider, so a credential that may set a person’s password, such as an admin’s console session or an admin API token with the identity area, can sign in as that person and, when the person holds an enroll role, enroll a phone or browser. | A device enrolled that way decides every request its person may decide, and each decision names that person. The straza.audit.admin record user.update with changed password names the credential that set the password, and the person’s own password stops working. |
Keep admin API tokens with the identity area few, give each the shortest lifetime the console offers, and revoke each one when its work is done. Run the enterprise profile, where your identity provider signs people in and the built-in issuer signs in only the break-glass admin. Review strazactl approvers list after a password change nobody expected. |
A confirmation on a separate device before a new approval device can decide. |
| A new approver certificate strands enrolled phones. A phone trusts the approver listener by the key it pinned at enrollment. | Losing or rotating the approver TLS pair makes every phone enroll again. | Back up the pair with the store, and compare the pin that /version reports after every move or restore. |
A rotated approver certificate that announces its successor over the authenticated channel. |
A database copy holds both signing seeds. The session and snapshot signing keys sit unencrypted in the signing_keys table, and the server publishes every snapshot key it stores, retired ones included. |
A copy mints any token for any user and signs a policy that every enrolled workstation accepts. No rotation withdraws an old snapshot key. | Protect the database and its backups as you would the keys. After a suspected copy, run strazactl signing-keys rotate session and revoke the device and approver credentials the old key signed, because it verifies them for 30 more days. |
Not on the roadmap. |
| Manifest values are stored in plain text. A server’s manifest is plain JSON in the database. Lists and exports mask secret-shaped values, and storage stays plain. | A credential typed into a manifest’s environment, arguments or address leaks with any copy of the database. A reader without apps:write can test a guessed value at a masked place through a drafts check. |
Keep every secret in the server’s secret, which strazactl apps secret set seals at rest. |
Not on the roadmap. |
| A Rego module can run past its time limit. The 100 ms limit is checked between evaluation steps, so one allowed built-in over a very large value runs to its end. The hook reads what the harness sends with no size cap. | A module can hold a large call for seconds and use memory in proportion before the decision denies. | Review a Rego module as code, and give the right to write policy only to people you would trust with code. | Not on the roadmap. |
A remote server may sit at an internal address. strazad refuses link-local, cloud metadata and unspecified addresses, and the enterprise profile refuses loopback unless apps.allowLoopbackUpstreams is true. Private addresses always pass. |
A manifest can point the gateway at any internal service that strazad’s network reaches. | Review the address of every remote manifest before you publish it, and limit strazad’s outbound traffic at the network. | An egress allowlist for each governed MCP server. |
| A manifest is trusted as written. Straza checks no publisher signature on a server manifest you import or write. | A server runs with the access and the credential you give it, whatever the source of its manifest. | Review each manifest before you publish it, and set admin.secondPerson so that a change that widens access needs a second person to publish. |
Manifests verified against a publisher signature before install. |
| Audit records can be lost. Server records wait in a memory queue of 4,096 before the database takes them, and the client spool drops its oldest records past 32 MiB while uploads fail, and reports the count to strazad. A record over 256 KiB is cut when strazad writes it. | A crash, a kill before the stop finishes, a full queue in standalone, or a failed write of an admin, identity, authentication or approval record loses records. The chain cannot show a record that never reached it. A cut record keeps only a prefix of its longest strings, each with its length and hash. One that cutting cannot fit keeps only who acted, on what, the outcome and reason, and the command, workspace and arguments, so its paths and other fields are lost. A row the event bus cannot carry, which happens only under an external NATS max_payload below 256 KiB or for a row written past strazad, is set aside and never reaches the chain. |
block, the enterprise default of governance.auditBackpressure, refuses a decision rather than run it without its record. Give strazad 30 seconds to stop, and watch straza_audit_lost_total, straza_audit_dropped_total, straza_outbox_set_aside_total and the straza.audit.tool records with the event audit.dropped, each a client’s report of what it lost. |
Not on the roadmap. |
| The hash chain cannot defeat the database owner. The chain proves that no record was edited or removed in place. | Someone who can write the database can rewrite every later record too, and a verify run inside that database finds the rewritten chain consistent. | A sink that keeps a copy of every record off the box, where that person cannot reach it. | Not on the roadmap. |
| Knowledge pack changes leave no audit record. Creating, binding, unbinding and deleting a pack writes nothing to the audit chain. | The chain cannot show who changed what an agent receives at check-in. | Grant identity:write, the area that covers packs, only to people and tokens you trust, and review strazactl packs list. |
Not on the roadmap. |
| Only the gateway checks objects. The hook’s decision on the agent’s machine never reads a covers line, and no policy rule can name an object. | A role that covers some repositories limits only the calls that go through the gateway. A server a person adds to their own harness has no object boundary. | Register every server whose objects matter behind the gateway, and remove the agents’ other routes to it. | Not on the roadmap. |
A tool whose calls do not name the object needs a role on every object. Straza does not check what a role’s own login reaches, so a tool with reach credential grants only through covers: every. |
On a server whose tools never name the object, such as a database server that runs every query under one login, no role can cover only some of its objects. | Register the server once for each narrower login, each with its own credential and roles. | Not on the roadmap. |
A role on every object covers objects made later. covers: every covers every object of the type, including objects no listing has seen. |
A repository created in the source system is open at once to every holder of such a role. | Give agents roles that name their objects. Set admin.serverAdminsCoverEvery to false, so only a global admin writes covers: every on a role a server owns. |
Not on the roadmap. |
| The object list is as fresh as the last listing. Straza lists a server every 10 minutes, and a refused call that names an object the list lacks asks for a listing at most once a minute per server. | A repository created or renamed since the last listing is refused to a role that names its objects, and after the first refused call it reaches the list in about two minutes when a call comes again after a minute, and otherwise at the next 10-minute listing. A person can name a new repository in a role only once a listing holds it. | For a renamed repository, refresh the list with strazactl apps objects refresh <server>, then try again. For a new one, refresh the list, then add the repository to the role’s covers line in a draft. A refresh is refused for a minute after a listing ran. |
Not on the roadmap. |
| An object deleted and made again under its old name. A role binds a named object to its id at publish, and the list maps the name to the old id until the next listing. | Until that listing, the gateway lets a role on the old object call the new one and sends the old id, so only a server that checks the id refuses the call. After it, the role covers nothing under that name until a person publishes it again. On a server whose listing gives no ids, the id is the value, so a role on the old object keeps covering the new one after every listing, until a person narrows the role. | Use a server that gives ids and refuses a call whose ai.straza/object-id is not the object’s current id. Publish the role again when it should cover the new object. On a server without ids, narrow every role that named a deleted object. |
Not on the roadmap. |
| The approval card shows the object only inside the arguments preview. An approval binds the object’s type and value, and the approval record has no object field of its own. | The preview is off when approval.preview.enabled is false, and past 2,048 bytes it cuts the middle of the arguments, so an approver can decide without seeing which object the call acts on. The approval still runs only on that object. |
Keep approval.preview.enabled true, the default, and do not approve a call whose preview leaves out the object. |
Not on the roadmap. |
| One object type per server and one argument per tool. A server pins at most one type, and a tool names its object in one required top-level string argument. | A tool that names its target in two arguments, such as an owner and a repository name, or in a nested or optional argument, cannot be kept to some objects. | Mark such a tool all, so only a role on every object calls it, and prefer servers that take the object as one string, such as Strazahq/pay-api. |
Not on the roadmap. |
| The object check rests on the server’s promises. Straza checks the value a call names and cannot see what the server does with it. | A server that follows a renamed object, acts on an object another argument names, or takes calls from outside the gateway breaks the boundary, whatever the roles say. | Mark a tool all unless its author makes the three promises, and let the server take calls from the gateway alone. |
Not on the roadmap. |
| An agent can ask its person to sign in for a key of its own. An agent that runs as the person can start a browser sign-in of its own, whose nonce names a key the agent keeps, and put the identity provider’s real sign-in page in front of the person. | If the person signs in, strazad enrolls the agent’s key as a machine of that person, because the sign-in is fresh and names that key. Straza asks the provider for a fresh sign-in each time, so a live sign-in cookie alone enrolls nothing where the provider honors that request. | Sign in only after you ran straza enroll yourself. Every enrollment writes a device.enrolled record, and the console lists each person’s machines, so review them and revoke a machine nobody recognizes. |
A confirmation on the person’s phone before a newly enrolled machine starts a session. |
Sandboxed commands on Windows cannot reach the service. Codex CLI and Gemini CLI run the commands of their sandboxes as another account or at a lower integrity level, and neither reaches straza service in this version. |
A Straza command that runs inside such a sandbox, such as straza exec, gets no session and stops. Nothing runs ungoverned, and the harness loses that command. |
Run Claude Code on Windows, whose commands run outside a sandbox, or run Codex CLI and Gemini CLI on Linux. | A way for those sandboxed commands to reach the service. |
| An agent inside WSL2 is not governed. Nothing connects WSL2 to the Windows service in this version. | An agent that a person starts inside WSL2 gets no decision and no session from the Windows service. | Run governed agents on Windows itself, and keep every credential an agent needs behind the gateway, where an ungoverned agent cannot reach it. | A way for agents inside WSL2 to reach the service. |
| The Windows binaries and the MSI are not signed. They carry no Authenticode signature, and cosign signs only the checksums file of a release. | SmartScreen may warn when a binary or the MSI runs for the first time, and Windows cannot tell by itself where the file came from. | Check the zip or the MSI with cosign and its checksum, as Install shows, before an administrator installs it. | Signed Windows binaries. |
| A power cut on Windows can revoke the machine. Microsoft does not document that the service’s write of a check-in counter survives a power cut. | When the last counter is lost, the next check-in repeats a number, and the server revokes the device as a copy, with a device.key.reused record although nobody copied the machine. Each person on it then enrolls again. |
Run straza enroll again on that machine. Before you treat a device.key.reused record of a Windows machine as a copy, ask whether it lost power. |
Not on the roadmap. |
| A force push through Straza looks like any other push. Straza and the git resource server see only commit ids, and a force push names the branch’s current commit as its old id, as a push that only adds commits does. | An approver who allows a push cannot see whether it rewrites the branch, and a role that may push may also force push. | A branch rule at GitHub on the branches you protect that blocks force pushes and that the App cannot bypass. | Not on the roadmap. |
Git is decided per tool, not per branch. A role or a rule names git_push, and no policy rule can name a branch yet. |
A holder of git_push on a repository may push to any branch of it that the resource server accepts. An approval still covers exactly the updates its card showed. |
Hold git_push for an approver with the default binding: call, and protect branches at GitHub. |
Not on the roadmap. |
The git host file is one per machine. straza service rewrites it from each check-in answer, whichever person’s session the answer belongs to. |
On a machine where people enroll with different Straza servers, the file follows the server that answered last, so a git host can stop going through Straza until the next answer, and git then reaches the host directly with no credential. | Enroll every person of a machine with the same server, which a managed install pins. | Not on the roadmap. |
Git through Straza is wired for Claude Code only. Only the Claude Code wiring points git at the host file, and a straza service started by hand with STRAZA_HOME writes its file where that wiring does not look. |
git in Codex CLI, Gemini CLI or beside a user-mode service goes to GitHub directly, with no credential, so a push or a private clone fails there. | Run Claude Code for agents that push, with the system service from the deb or rpm package on Linux, the Straza package for macOS on a Mac or the Straza MSI on Windows. | Codex CLI and Gemini CLI get no git wiring yet. |
| The Straza package for macOS is signed only from release 2.0.0 on. Those releases carry it as a pkg signed with the Developer ID of SynapTech s. r. o. and notarized by Apple. Releases up to 1.1.2 carry none, and a package that an administrator builds from the source tree without a Developer ID is unsigned and not notarized. | On a Mac with such a build, nothing outside your own build shows where the package came from, and macOS cannot check its origin. | Install the release pkg, checked with cosign and its checksum and on a Mac with pkgutil and spctl, as Install shows. Build from source only on a Mac you control, or sign and notarize the build with Developer ID identities of your own through the build script’s signing variables. |
Not on the roadmap. |
| The service on macOS can be switched off in System Settings. On a Mac without a Managed Login Items profile for the team identifier WMPWM57KYK, macOS lists the service under Login Items & Extensions, as SynapTech s. r. o. for the release pkg, where it can be switched off. Only a device management tool can install that profile, on macOS 13 or later. | While it is off, every tool call the hooks govern on that Mac is refused. Nothing runs ungoverned, and the agents there stop working. | The profile that Deploy with Intune and Deploy with Jamf Pro install, which no tenant has run yet. When the hooks say that the background service isn’t running, sudo launchctl print system/ai.straza.service shows whether it runs, and Login Items & Extensions shows whether it was switched off. |
Not on the roadmap. |
| Claude Code skips Straza’s hooks under a higher managed source. Claude Code uses only the highest of its managed sources. On a Mac, a Claude Code profile from a device management tool ranks above the managed settings file that carries Straza’s hooks, as do server-managed settings on every system and a registry policy on Windows. | On a machine with such a source, Claude Code runs without Straza’s hooks and gives no warning. | Ship no Claude Code profile, server-managed settings or registry policy to the machines whose agents you govern, and keep every credential an agent needs behind the gateway. | Not on the roadmap. |
Sandboxed shell commands on macOS cannot reach the service. When an administrator turns on Claude Code’s sandbox for shell commands, a command in it reaches only the Unix sockets that the sandbox’s allowUnixSockets setting names, and the wiring does not name the service’s socket. |
git through Straza fails in such a command, and so does any other Straza command run there, such as straza exec. Nothing runs ungoverned, and the agent loses that command. |
Leave Claude Code’s sandbox for shell commands off on the Macs whose agents use git through Straza. | Not on the roadmap. |
A hook that cannot run lets the call through on some harnesses. Codex CLI, Gemini CLI and Claude Code releases before 2.1.295 run a tool call when straza hook cannot start, is killed or times out, because they offer no setting that blocks the call instead. |
On those harnesses a missing or broken Straza binary leaves the agent ungoverned, and its calls leave no Straza audit record. | Run Claude Code 2.1.295 or later, and run straza install --managed again after the server upgrade, so the wiring carries the setting that blocks such a call. Remove the wiring with straza uninstall --managed before removing the package. |
Not on the roadmap. |
A few settings move these limits when you change them from their defaults, and Hardening lists every security setting with its default in both profiles.
Scale ceilings
One deployment of Straza also has ceilings on how much it carries. Each row gives the ceiling, the reason it exists and what to do as you approach it. Requirements and sizing gives the measured runs behind the numbers.
| Ceiling | Why it is there | What to do near it |
|---|---|---|
| A standalone server runs on one host, as one process. | It keeps its SQLite store and its embedded event bus in its own data directory, so a second replica would hold a second, separate store. The Helm chart refuses to render the standalone profile with more than one replica. | Move to the enterprise shape when you need a second replica. |
| 1,000 connected agents per strazad is the most any capped run held. | The run passed every check in 2 CPUs and 1 GB on SQLite, and no capped run went higher. | Above 1,000, give strazad more memory and processors or add enterprise replicas, and measure your own load. |
| The server’s audit queue holds 4,096 records in memory. | Each record waits there between its decision and the database write. | In standalone a full queue drops records and counts them in straza_audit_dropped_total. In enterprise, under block, a decision waits up to 25 seconds for room and is then refused. Keep the database reachable. |
| The chart’s 512 MiB limit is reached above about 250 server decisions a second while the database is down. | Under block, each waiting decision holds memory and its client’s request for up to 25 seconds. |
Raise the memory limit above your peak decision rate, or spread the load over more replicas. |
| A replica opens up to 16 Postgres connections, or 4 for each processor Go uses when that is more. | strazad bounds its connections, so a burst waits inside strazad instead of exhausting the database. | Keep Postgres max_connections, 100 by default, above 16 for each replica under the chart’s CPU limit of 1, plus your other clients. The chart’s autoscaler, when you turn it on, allows 10 replicas by default. |
| The client spool holds up to 32 MiB of parked records, each under 3 MiB. | It bounds the disk a client uses while strazad cannot be reached. | Once an upload failed, past the cap the oldest unsent records are dropped. straza doctor reports the loss, and the next upload reports it to strazad as an audit.dropped record. Keep clients able to reach strazad. |
| The client’s background service works on at most 16 decide and capture calls and 17 MiB of their bodies for each account at once, and keeps at most 64 KiB of a recorded command and of its paths. | It bounds the memory one account, an agent’s included, can make the shared service hold. | A call past the bound is refused, a decide with a sentence that asks for one more try in a minute, and counted in the next audit.dropped record. A cut record keeps the SHA-256 of the whole command or path list. |
| 65,536 clients can hold the push stream open at one strazad. | events.pushEdgeMaxConns bounds the open push streams of one process. |
A client over the cap polls instead until a retry gets in. Raise the setting or add replicas. |
| The audit stream on NATS keeps 60 days and at most 2 GiB by default. | Unbounded, a full disk would stop strazad from publishing any event, kill-switch revocations included. | The stream drops its oldest records first. Keep events.auditStreamMaxBytes well below the NATS volume, 5Gi in the chart. |
A server manifest’s limits.cpu and limits.mem are stored and not enforced. |
strazad checks their format and nothing reads them yet. | Bound the server’s processors and memory in the runtime that runs it, such as its container or pod. |
The roadmap plans cells, independent deployments with their own database and event spine, as the unit of growth past one deployment.
Report anything that contradicts this page through Reporting a vulnerability.