← All tools · Guides

Watchlist alerts: a big move or a volume spike, pushed once

Local review series, part 4. The rules watch the list. You look only when one of them fires. This closes the local path.

How to choose: An alert is there so you can stop watching the screen. The previous note already has the Python environment and a push channel, ServerChan or email. Reuse both. The only design rule is to miss a move before you flood the phone. Start with two or three rules, run them for a week, and add more only after you still read the ones that arrive. During the China A-share session, call ak.stock_zh_a_spot_em once a minute, keep the rows whose code is on your watchlist, and rename the Chinese columns before you compare anything. Do not shorten the sleep to a second. That is the fastest way to get the IP blocked. Write the thresholds in rules.yaml, not in the script, so a change does not touch code. The same name and the same rule push once per day. An in-memory set of (code, rule) is enough. A restart can send one extra message, and that is acceptable. Where the loop runs is a separate choice. Leave the PC on and start the script at 09:25, stopping at 15:10, if you are already at the machine. Move the same script to a cloud server, under nohup or systemd, if the laptop lid will be shut. If you do not want a loop all day, run the same rules at the open, at 14:30, and at the close. The review note, the cloud server note, and TradingView alerts are different articles. This page is not the tool-order guide and not the site FAQ.

Who it is for: Someone who already has the local review stack, watches a short China A-share list during the session, and wants a phone push for a large rise, a large drop, or a volume spike without staring at quotes. What you need from this page is the one-minute poll, two or three rules, the once-a-day throttle, and a decision about whether the PC stays on.

Who it is not for: Someone who wants a broker terminal, a Level-2 feed, or an alert that places an order. This page only notices. It does not choose the trade. Chart alerts on a TradingView layout are a different article. A machine that must stay on after the close starts with the cloud server note, which is also a different article. This page is not the after-close review, not the tool-order guide, and not the site FAQ. Do not treat a push as a reason to buy or sell.

The same walkthrough in Chinese is a watchlist alert for a large move or a volume spike. A machine that stays on, including the first SSH login, is opening a cloud server. The same article in Chinese is the Chinese cloud server note. That note is not this article. Alerts drawn on a chart, on the free plan, are the TradingView setup. The same article in Chinese is the Chinese TradingView note. That article is not this one either.

Facts below follow the Chinese walkthrough. The commands were not re-run for this English page. This is not investment advice. Free data and push quotas follow the vendor's current page.

01 · The point is to look less

The failure mode of an alert is fatigue. Loose rules send dozens of messages, and by the third day you ignore the phone. Prefer a miss to a flood. Open with the two or three moves that would still pull you away from other work, keep them for a week, and only then add a rule. A new high or a new low belongs in that later pass, once you have a reference price to compare. This page does not invent that level for you.

02 · Live quotes: one snapshot a minute

import akshare as ak
import time
from datetime import datetime

def market_open() -> bool:
    now = datetime.now()
    t = now.hour * 100 + now.minute
    return now.weekday() < 5 and (925 <= t <= 1130 or 1300 <= t <= 1505)

while market_open():
    try:
        spot = ak.stock_zh_a_spot_em()          # one snapshot of the whole market
        rows = spot[spot["代码"].isin(WATCHLIST)]
        check_rules(rows)                        # section 03
    except Exception as e:
        log(f"fetch fail: {e}")                  # log it, then the next round
    time.sleep(60)                               # once a minute is enough

The snapshot uses Chinese column names. Rename them before the rules run. The five this page needs are code, name, last price, percent change, and volume:

rows = rows.rename(columns={
    "代码": "code",
    "名称": "name",
    "最新价": "price",
    "涨跌幅": "pct",
    "成交量": "volume",
})

WATCHLIST is the short list you already watch. One request a minute is the whole poll. Leave the 60-second sleep alone.

03 · Rules in a file, and a once-a-day throttle

Thresholds live in the config. Changing a number should not mean editing the script:

# rules.yaml
- name: large rise
  cond: "pct >= 5"
- name: large drop
  cond: "pct <= -5"
- name: volume spike
  cond: "pct >= 3 and vol_ratio >= 2"   # live volume / 5-day average, scaled by time open

pct is already in percent points. A print of 5.1 is a rise of 5.1 percent. Compare it with 5. vol_ratio is not a column in the snapshot. Compute it from volume and the recent average, scaled by how much of the session has elapsed, before the rule runs.

The throttle is what keeps the phone quiet. The same name and the same rule go out once:

seen: set[tuple[str, str]] = set()

def already_sent(code: str, rule: str) -> bool:
    key = (code, rule)
    if key in seen:
        return True
    seen.add(key)
    return False

The set is in memory. A restart drops it, and the next fire can push one extra time. That is acceptable. The message is three lines: code and name, the rule that fired, and the price with the percent. Send it with the ServerChan or email function from the review note.

04 · Where the computer sits

Pick one of these. They share the same rules.

• The PC stays on. This is the default. You are at the machine during the session. Task Scheduler starts the script at 09:25. market_open ends the loop at 15:10.
• A cloud server. The laptop lid will be shut, and you still want the minute loop. Copy the script and run it under nohup or systemd. Opening the machine is the cloud server note.
• Three clocks, no loop. Run the snapshot at the open, at 14:30, and at the close. You give up some of the timeliness. The cost is almost nothing.

05 · Acceptance

• Find a watchlist name near a threshold, widen that threshold for one trial, and confirm the phone receives the push.
• Fire the same name and the same rule again. It must not push a second time.
• Drop the network for one round. The script logs the error and stays up. Restore the network and the next round runs.

06 · Pitfalls

• The percent is already a percent. Compare pct with 5. Dividing by 100 makes every rule miss.
• A halted name. Volume is 0 and the price sits at the previous close. That can fire a junk alert. Exclude volume == 0.
• The call auction. Prints before 09:25 jump around. Turn the rules on after the open.
• A push is not a decision. What you do next comes from the review and the plan. The alert helps you see the move. It does not choose the trade.