This page collects every limit that is fixed in the code and everything ramzes does not do. Nothing
here can be changed by configuration unless the row says so.
- Several instances working together. ramzes runs as one instance. There is no high
availability and no horizontal scaling. Noise-control decisions are made under a lock in one
process, and the ingest rate count and sign-in lockout are kept in that process’s memory.
- Roles. Every user and every API token can read and change everything.
- Single sign-on. Sign-in is email and password. No OIDC, SAML or LDAP, and no second factor.
- An audit log. The configuration history records who changed the configuration. Sign-ins,
reads, and changes to users and API tokens are not recorded.
- Password change or reset. A password is replaced by removing the user and adding it again.
- Full-text search. The feed filters by time, source, topic, minimum severity, one label,
outcome and a substring of the title, and sorts by arrival, severity, topic or title.
- A memory limit for templates and expressions. They are stopped after a fixed amount of work,
but a signed-in user can still write one that builds a very large string.
- TLS. ramzes serves plain HTTP. Put it behind a reverse proxy.
- CORS. No cross-origin headers are sent, so a web page on another origin cannot call ramzes.
- Publish/subscribe. Services cannot read events back or subscribe to topics. ramzes is a
router, not a broker.
- On-call schedules. An escalation chain names connections and delays. There are no rotas or
calendars.
- End-user notifications. No subscriber lists, preferences or unsubscribe handling.
- More than one workspace. An instance has a single workspace; there are no tenants or teams.
- A hosted version, a published image, a chart repository or binary downloads. You build from a
checkout.
- A backup command for Postgres.
ramzes backup covers SQLite. For Postgres use pg_dump.
- A “recovered” event from ramzes itself.
ramzes.connection.paused and
ramzes.delivery.failed are raised as firing only.
| Limit |
Value |
| Request body |
1 MiB (1,048,576 bytes) |
| Events in one request |
1000, both for a batch to /v1/events and for what a mapping produces from one request |
| Events per source per minute |
6000; beyond that 429 with Retry-After |
topic |
1 to 200 characters: letters, digits and . _ : / - |
title |
1 to 500 characters, not blank |
body |
up to 20,000 characters |
labels |
up to 64 |
| Label key |
1 to 100 characters: letters, digits and . _ - |
| Label value |
up to 1000 characters |
dedup_key |
1 to 200 characters |
Idempotency-Key header |
1 to 200 visible ASCII characters |
severity |
one of debug, info, warning, error, critical |
payload |
any JSON, within the size of the request |
| Unknown fields |
refused |
| NUL characters |
refused in title, body, label values and dedup_key |
| Source name |
1 to 100 characters |
| A source key you supply yourself |
16 to 200 characters |
| Recorded requests per source |
the newest 20, bodies cut to 10,000 characters (/v1/events) or 64,000 (per-source addresses) |
| Age of a signed request (GitLab signing token) |
5 minutes |
The ingest rate limit is counted per source and clock minute, in memory.
| Limit |
Value |
| Names of routes, templates, connections, connectors, sources and API tokens |
1 to 100 characters |
| Targets of a route |
at least 1 |
| Dedup window |
1 second to 30 days |
| Dedup key computed by an expression |
cut to 200 characters |
| Group wait |
1 second to 1 day |
| “Send again while still open” interval |
1 minute to 30 days |
| Times one notification is sent to its route’s target |
20, counting the first message and every reminder; then the reminders stop. Escalation steps and the message that says it is resolved are separate. |
| Rate limit |
at least 1 notification per window; window 1 second to 1 day |
| Quiet hours |
one daily period per route; start and end must differ |
| Titles listed in a digest of several events |
10, then “and N more” |
| Escalation steps |
up to 10 |
| Delay of an escalation step |
1 second to 7 days |
| Length of a silence |
up to 366 days; it needs at least one condition |
| Comment on a silence |
500 characters |
| A silence made from a Slack button |
1 hour, on the topic and up to 10 labels of the first event |
| Compiled size of a label regular expression |
100,000 bytes |
| Limit |
Value |
| Work per evaluation |
50,000 instructions |
| Source text of one expression or template |
32 KiB |
| Output of a template |
256 KiB |
| Memory |
no limit |
| Limit |
Value |
| Time for one request to a channel |
10 seconds |
| Redirects |
never followed |
| Answer read from a service |
64 KiB; 500 characters are kept as the error text |
| Retry schedule |
5 seconds, doubling to at most 1 hour, each delay varied by up to 20 % either way |
| Attempts per delivery |
8 |
| Longest a delivery is retried |
24 hours from when it was queued |
| Statuses that are retried |
408, 425, 429, 5xx, timeouts and network errors. Any other non-success is final. |
| Deliveries sent at the same time |
16 in all, 4 per connection |
| Failures before a connection is paused |
5 in a row |
| Length of a pause |
1 minute, doubling to at most 15 minutes |
| Guarantee |
at-least-once: a message can arrive more than once |
Each channel also cuts long messages to what its service accepts. See the
channel pages.
| Limit |
Value |
| Event retention |
30 days by default; 0 (forever) to 3650 days; set under Settings |
| Per-topic retention |
exact topic names only, no patterns |
| Retention pass |
at start and every hour |
| Configuration history |
the newest 200 entries |
| Failures listed |
the newest 200 |
| Databases |
SQLite (a file in the data directory) or Postgres |
| Limit |
Value |
| Password |
at least 10 characters |
| Email address |
up to 254 characters |
| Session |
30 days from sign-in, not extended by use |
| Sign-in lockout |
10 wrong passwords per email in 15 minutes; counted in memory for up to 10,000 addresses |
| API tokens |
no expiry; revoked by hand |
| Page size of the feed and of notifications over the API |
1 to 200, default 50 |
| Period of the overview |
1 to 744 hours (31 days) |
| Connection health shown |
the last 24 hours |
| Other rate limits on the admin API |
none |
| Limit |
Value |
| Instances |
1 |
/metrics |
open unless a metrics token is set |
| Metric labels |
connector kind and outcome only |
| Time given to deliveries in flight at shutdown |
20 seconds |
| Secret key |
32 bytes; if it is lost, stored credentials cannot be recovered |
| Acknowledgement links |
no expiry; valid while the notification is kept and the secret key is unchanged |
